Attitudes Towards Corruption and Their Consequences on Political Behavior
ADVERTIMENT. Lʼaccés als continguts dʼaquesta tesi queda condicionat a lʼacceptació de les condicions dʼús establertes per la següent llicència Creative Commons: http://cat.creativecommons.org/?page_id=184
ADVERTENCIA. El acceso a los contenidos de esta tesis queda condicionado a la aceptación de las condiciones de uso establecidas por la siguiente licencia Creative Commons: http://es.creativecommons.org/blog/licencias/
WARNING. The access to the contents of this doctoral thesis it is limited to the acceptance of the use conditions set by the following Creative Commons license: https://creativecommons.org/licenses/?lang=en
Attitudes towards corruption and their consequences on political behavior
Sofia Breitenstein Gomis
PhD Thesis Politics, Policies and International Relations Department of Political Science and Public Law
July 2020
Supervisors: Dr. Eva Anduiza Perea Universitat Autònoma de Barcelona
Dr. Jordi Muñoz Mendoza Universitat de Barcelona
Para Joan, nuestra hija y las/los que están por venir
Table of contents
ACKNOWLEDGMENTS ...... 8 ABSTRACT ...... 12 FIGURES AND TABLES ...... 15
FIGURES ...... 15 TABLES ...... 17 CHAPTER 1 INTRODUCTION ...... 19
1.1 THE PUZZLE...... 20 1.2 EXISTING EXPLANATIONS ...... 25 1.2.1 The ignorant voter ...... 25 1.2.2 The indifferent voter ...... 28 1.3 OVERVIEW OF THE ARGUMENT: THE MULTIDIMENSIONAL VOTER ...... 33 1.4 CONTRIBUTION TO THE TRADEOFF ARGUMENT ...... 36 1.5 RESEARCH DESIGN ...... 39 1.5.1 The Spanish case ...... 39 1.5.2 Data and methods ...... 44 1.6 OUTLINE OF THE DISSERTATION ...... 50 CHAPTER 2 SOCIAL DESIRABILITY BIAS ...... 52
2.1 INTRODUCTION ...... 52 2.2 SOCIAL DESIRABILITY IN ATTITUDES TOWARDS CORRUPT POLITICIANS ...... 54 2.3 THE TRADEOFF ARGUMENT ...... 55 2.4 EMPIRICAL STRATEGY: SOCIAL DESIRABILITY BIAS ...... 56 2.5 RESULTS: SOCIAL DESIRABILITY BIAS ...... 60 2.6 EMPIRICAL STRATEGY: TRADEOFF ARGUMENT ...... 62 2.7 RESULTS: TRADEOFF ARGUMENT...... 63 2.8 CONCLUSIONS ...... 65 CHAPTER 3 TRADEOFFS AND CORRUPTION...... 66
3.1 INTRODUCTION ...... 66 3.2 THEORETICAL FRAMEWORK ...... 67 3.3 EMPIRICAL STRATEGY ...... 71 3.3.1 Design and wording of the experiment ...... 74 3.4 RESULTS ...... 78 3.5 CONCLUSIONS ...... 82
CHAPTER 4 REASSESSING THE COMPETENCE CORRUPTION TRADEOFF ...... 84
4.1 INTRODUCTION ...... 84 4.2 THEORETICAL FRAMEWORK: ...... 87 4.2.1 Outcome accountability versus procedural accountability ...... 87 4.2.2 The mechanism: malfeasance, politicians’ traits, and voting decisions ...... 90 4.3 EMPIRICAL STRATEGY ...... 94 4.4 RESULTS ...... 99 4.4.1 Average treatment effects ...... 99 4.4.2 Mediation analysis ...... 102 4.5 CONCLUSIONS ...... 107 CHAPTER 5 WHO CARES? INDIVIDUAL HETEROGENEITY IN THE PUNISHMENT OF CORRUPTION ...... 109
5.1 INTRODUCTION ...... 109 5.2 THEORETICAL FRAMEWORK ...... 112 5.2.1 Gender ...... 112 5.2.2 Political sophistication ...... 114 5.2.3 Age ...... 117 5.3 EMPIRICAL STRATEGY ...... 120 5.4 RESULTS ...... 122 5.5 CONCLUSIONS ...... 129 CHAPTER 6 WHAT ABOUT THE OTHER ONE? CHARACTERISTICS OF THE ALTERNATIVE CANDIDATES ...... 131
6.1 INTRODUCTION ...... 131 6.2 THEORETICAL FRAMEWORK ...... 133 6.3 EMPIRICAL STRATEGY ...... 135 6.4 RESULTS ...... 137 6.5 CONCLUSIONS ...... 139 CHAPTER 7 CONCLUSIONS ...... 141
7.1 OVERVIEW OF MAIN FINDINGS ...... 141 7.2 IMPLICATIONS OF MAIN FINDINGS FOR ACCOUNTABILITY ...... 146 7.3 IMPLICATIONS FOR ANTI-CORRUPTION CAMPAIGNS ...... 149 7.4 CONCLUDING REMARKS ...... 151 APPENDICES ...... 152
APPENDIX A: SUPPLEMENTARY MATERIALS FOR CHAPTER 2 ...... 152 APPENDIX B: SUPPLEMENTARY MATERIAL FOR CHAPTER 3 ...... 155 B.1 Robustness check of relative weight hypothesis ...... 156
B.2 Robustness check of the conditional punishment hypothesis ...... 158 B.3 Conditional punishment of gender, education and management experience161 B.4 Conditional punishment disentangling corruption accusations ...... 162 B.5 Further robustness checks ...... 163 B.6 Difference with other studies using conjoint experiments ...... 171 B.7 Questionnaire ...... 173 APPENDIX C: SUPPLEMENTARY MATERIAL FOR CHAPTER 4 ...... 176 C.1 Manipulation checks ...... 177 C.2 Additional mediation analyses ...... 178 C.3 Mediation sensitivity analyses ...... 179 C.4. Average treatment effects including non-partisan attachment ...... 181 BIBLIOGRAPHY ...... 183
Acknowledgments
El largo proceso hasta la culminación de una tesis doctoral requiere de la ayuda de personas generosas, pacientes y sabias. Yo he tenido la suerte de contar con muchas, tanto en el entorno académico como en mi vida privada, que me han facilitado llegar hasta aquí.
Las personas más importantes en este proceso han sido, sin duda, mis supervisores: Eva Anduiza y Jordi Muñoz. Me siento extremadamente afortunada de haberos tenido como mentores durante estos años, vuestro apoyo y confianza han sido imprescindibles para llegar hoy a entregar esta tesis doctoral. Gracias por enseñarme todo lo necesario para realizar la tesis y transmitirme vuestra pasión por la investigación en ciencias sociales. Gracias Eva por tu inmensa generosidad y empatía, por compartir tus vastos conocimientos y por acogerme en tu grupo y proyectos de investigación. Gràcies Jordi per ensenyar-me cada dia una cosa nova, per la teva atenció als detalls i per sempre animar-me a apuntar més alt. Gracias a ambos por vuestra creatividad y rigurosidad en la investigación, por vuestra enorme capacidad de enseñar y generosidad sin límites. Sois un ejemplo a seguir tanto a nivel profesional como personal.
Mis compañeros del grupo de investigación DEC también han tenido un papel importante en el desarrollo de esta tesis. Gracias a todos por vuestros consejos de como “sobrevivir” en la academia, por las discusiones en las que he aprendido los aspectos prácticos de la investigación, por las comidas en el rectorat y las risas en las conferencias y cenas de navidad. Gracias Quique por ser un referente académico del que he aprendido mucho, pero a la vez un buen amigo con el que
8 Acknowledgments
me he reído mucho. Por confiar en mí para escribir artículos juntos y por ofrecer tu hombro en los momentos más difíciles. Gracias Maca por también confiar en mí e invitarme a escribir con vosotros lo que sería mi primer trabajo publicado. Y por volver a contar conmigo para realizar tu maravillosa idea. Gracias Roberto por guiarme en el inicio del doctorado. Por tu apoyo en mi primera conferencia, ensayar conmigo la presentación, pero sobre todo por no dejarme cometer el error de comer pizza en Roma (!) Gracias a Guillem (Rico) por criticar cada uno de los experimentos que he diseñado, aunque muchas veces no te gustaran mis propuestas, he aprendido un montón de todos tus comentarios. Gracias a Berta por hablar abiertamente de las dificultades emocionales de hacer una tesis. Te admiro por tu valentía. Gracias Carol por discutir al más mínimo detalle los papers, he aprendido mucho escuchándote. También gracias a Carol y Marc por la pasión que le ponéis en todo lo que hacéis. Gracias Jordi G. por siempre estar dispuesto a ayudar, por compartir penas y consejos de edición de tesis. Gracias Camilo por ayudarme con el conjoint y por compartir do-files que aún sigo ojeando. Gracias a todos los que habéis hecho que el doctorado a veces sea un placer: María José, Lucia, Sheila, Dani M., Juan, Luca, Roser, Dani B., Sabina y Ángel. Y gracias a Eva por dirigir este fantástico grupo, crear un ambiente de trabajo generoso y colaborativo en un contexto tan difícil como es la academia. Será difícil volver a encontrar un entorno de trabajo en el que me sienta tan cómoda.
També molts gràcies als meus companys de doctorat, Guillem (Ripoll) i Marina amb els que ha sigut més fàcil i divertida aquesta tasca. Us desitjo tot el millor!
A very special thanks to Michael Lewis-Beck for believing in me, including me in the authenticity project and for always sending supporting words along. It is seldom that one finds such a generous and kind person in the academia.
Gracias a Víctor Lapuente por ser mi mentor en la Universidad de Gotemburgo, por tus sabios consejos y simpatía.
Gracias a Pilar Sorribas por darme la oportunidad de participar en el proyecto “LIMCOR: Limits to political corruption”.
A big thanks to Alice and Oul for being such inspiring young female academics and good friends. Also, thanks to the other EITM friends for the fantastic time we spent together: Marta, Jonas, Matteo and Sanja.
9 Acknowledgments
Also a special thanks to Dieter, John, Jac and Florence for being such a great team. Working with you has been a pleasure.
A big thanks to Alexandra Elbakyan and collaborators for creating Sci-hub, without which it would have been impossible to access all the research used to write this dissertation. Thanks for breaking the barriers to knowledge.
I am also grateful to the FPI grant (BES-2015-072756) from the Spanish Ministry of Economy and Competitivity and the European Social Fund, which funded my PhD studies. And to several projects which allowed me to gather the data used in this dissertation and to disseminate my findings: “Populist Attitudes in Spanish Public Opinion”, funded by the Spanish Ministry of Economy and Competitiveness (CSO2014-52950-R). “Living with Hard Times: How Citizens React to Economic Crises and Their Social and Political Consequences” (LIVEWHAT), project funded by the European Commission under the 7th Framework Program (grant number 613237). “LIMCOR: Limits to political corruption” (Fundació La Caixa 2016 ACUPO177). “Political Change in Spain: Populism, Feminism and new dimensions of conflict” (CSO2017-83086-R).
Finally, a special thanks to all the woman in research that have inspired me: Eva Anduiza, Elvira Carrió, Macarena Ares, Alice Iannantuoni, Marga León, Carol Galais, Oul Han, Cristina Carrasco, Dani Marinova, Berta Barbet, Laia Montoliu- Gaya, Elena Layunta, María José Hierro, Sheila González, Lisanne de Blok, Elena Costas-Pérez, Sandra León, Marta Fraile, Eva Østergaard-Nielsen, Aina Gallego, Nara Pavão, Catherine De Vries, Anja Neundorf, Pilar Sorribas, Lena Wängnerud, Esther Duflo and many more.
10 Acknowledgments
También quiero agradecer el apoyo de mis amigos y familia, sin los que estos años habrían sido mucho más difíciles.
Gracias a Joan por ser mi compañero de vida, por tu alegría y siempre conseguir sacarme una sonrisa y por tu capacidad de restar importancia a todo. Sin tu apoyo no habría podido llegar hasta aquí.
Gracias a Elvira y Tamara por estar en mi vida y por compartir tantos momentos inolvidables. Vuestro apoyo ha sido fundamental para llegar hasta aquí. Gracias Elvira por haber compartido tantas vivencias y por inspirarme en este camino. Gracias Tamara por venirme a ver cuando más lo necesitaba.
Gracias a Lau por ser la primera persona en hablarme del apasionante mundo de la sociología. Sin tu visita guiada a la facultad de la UB seguramente mi vida habría tomado otro rumbo. Gracias Sergio por tu compresión y apoyo. Gracias a los dos por contagiarme vuestra pasión por las ciencias sociales. Gracias a las Burgers por vuestros debates interminables que han despertado mi interés en la política y me han empujado a siempre querer aprender más. Gracias por tantos años de amistad y por ser una gran familia.
Gracias a Olivia y Diego por haberme hecha tía de las sobrinas más monas del mundo. Ver crecer a Luna, Mauro, Andrea y Alba ha sido uno de los mayores placeres de mi vida. Gracias a mis padres por darme absoluta libertad en escoger mi camino. Gracias Moma por haberme preguntado en cada visita cuánto me queda para ser doctora y por haber sido una mujer tan valiente. Gracias a Bea por tus consejos y apoyo emocional.
Gracias a las Truchas por tantas horas de diversión. Vuestra alegría y creatividad hace que cualquier día pueda ser una aventura. Gracias a Tomek por tantos años de amistad y por venirme a ver cuando más lo necesitaba. Gracias a Laia y Elena por no dejar que me rindiera. Gracias a Sonia, Mireia, Cris, Elena y Elvira por ser mis primeras amigas en Barcelona.
11
Abstract
Although corruption has severe negative consequences for society, the reelection of corrupt governments or politicians is all too frequent. Vehement condemnation of corruption is widespread, while at the same time, malfeasant behavior has rather limited consequences at the polls. This situation still represents a paradox in the social sciences. According to standard democratic theory, elections are expected to serve as an instrument to hold politicians to account; however, studies conducted in multiple countries indicate that voters’ punishment of malfeasant politicians is rather limited. The aim of this thesis is to assess citizens’ attitudes towards corruption and its relative importance on their voting intentions, in order to provide a better understanding of corruption accountability.
A recurring explanation for why voters do not sanction corruption more severely is that they are either insufficiently informed about the wrongdoings (ignorant voters) or that they are actually not that worried about malfeasance (indifferent voters). This dissertation provides compelling evidence that voters do indeed care about corruption and that ideally, they would like to punish the corrupt politician. Nevertheless, holding politicians to account is not as simple as it may seem. Besides integrity, voters consider many other important aspects when casting a ballot. Voting is a multidimensional decision and electors may trade integrity against other characteristics that they value.
In line with the indifferent voter argument, an interpretation of the overwhelming disapproval of corruption in surveys is that these answers are plagued with social desirability bias. According to this position, the rejection of corruption that
12 Abstract
citizens express in surveys is driven by their will to express socially accepted attitudes. By using an original list experiment, Chapter 2 of this dissertation shows that respondents’ intentions to vote for a corrupt candidate from their preferred party does not increase when the question is formulated in an unobtrusive way, even though this technique is specifically designed to increase respondents’ willingness to express their truthful positions. However, a respondent’s intention to vote for the corrupt politician does increase when the question is formulated as a tradeoff. Therefore, the main problem of standard survey questions that ask about attitudes towards corruption is not social desirability bias, but their inability to replicate the multidimensionality of real elections.
Keeping in mind the complexity of making decision in elections, Chapter 3 uses a conjoint experiment to reflect this multidimensional scenario and to thus measure the relative importance of corruption on voting intention. This chapter provides clear-cut evidence that, when faced with a multidimensional decision, voters are indeed willing to trade off corruption for other valued characteristics such as partisan identity or economic performance. The results show that co- partisanship determines voting choice to the same extent as corruption. Moreover, both co-partisanship and, to some extent, economic performance, moderate the negative effect corruption has on the vote.
Besides focusing on the tradeoffs that voters face when casting a vote, this dissertation also aims to increase our understanding of the tradeoff argument. This has been carried out by (i) identifying the causal mechanisms that lead voters to (not) vote for a malfeasant politician (ii) exploring what individual characteristics of voters increase the probability of them trading integrity against representation or competence and (iii) examining what characteristics of the alternative candidates cause the punishment of the corrupt politician to be increased further. Chapter 4 shows that a drop in the level of trust felt towards the corrupt politician in question explains why voters may decide not to vote for her. Chapter 5 identifies some modest but potentially relevant heterogeneities in citizens’ responses to corruption, while Chapter 6 shows that voters punish the corrupt politician by switching to the alternative option when this is an attractive candidate (one who has a strong economic performance or belongs to a party they have a certain predisposition to vote for).
13 Abstract
All in all, this dissertation provides compelling evidence that voters value politicians’ integrity and they would like to elect honest governments. But in elections, they may trade off integrity for other valued characteristics. Consequently, the results show that some of the failures of accountability regarding corruption are not due to voters’ ignorance or their indifference towards politicians’ wrongdoings, but because integrity is not the only aspect that matters in their decision making. While elections may be the best instrument for selecting those who govern, this dissertation also points out the limitations of elections as accountability mechanisms.
14
Figures and Tables
Figures
Figure 1.1. Correlation between Level of Democracy and Corruption Perception Index 23
Figure 1.2. Percentage of respondents that consider that if “a government official gives a job to someone from his family who does not have adequate qualifications”, this is wrong and punishable, wrong but understandable, or not wrong at all 29
Figure 1.3. Percentage of respondents that consider that if “a government official demands a favor or an additional payment for some service that is part of his job”, this is wrong and punishable, wrong but understandable, or not wrong at all 30
Figure 1.4. Percentage of respondents that consider that if “a public official decides to locate a development project in an area where his friends and supporters live”, this is wrong and punishable, wrong but understandable, or not wrong at all 31
Figure 1.5. Percentage of respondents that consider that it is never justifiable to accept a bribe. (1: it is never justifiable and 10: it is always justifiable) 32
15 Figures and Tables
Figure 1.6. Corruption Perception Index 2018 41
Figure 1.7. Percentage of respondents that consider that corruption is the most important problem Spain is facing 42
Figure 1.8. Percentage of respondents (households in European countries) that paid a bribe in 2017 across different countries and regions 44
Figure 3.1. Average marginal component effects (AMCE) 80
Figure 4.1. Analytical framework 91
Figure 4.2. Mean propensity to vote across different treatment conditions 99
Figure 4.3. Analyses of traits: proportion of respondents that consider the mayor trustworthy and efficient across different treatments conditions 100
Figure 4.4. Mediation analysis of trustworthiness with confounding alternative mechanism (efficiency) 104
Figure 4.5. Mediation analysis of efficiency with confounding alternative mechanism (trustworthiness) 105
Figure 4.6. Mediation analysis of empathy with confounding alternative mechanism (efficiency) 106
Figure 5.1. Comparison of marginal means for men (1) and women (1) 124
Figure 5.2. Comparison of marginal means for low sophisticates (1), average (2) and high sophisticates (3) 124
Figure 5.3. Comparison of marginal means for respondents who are 18 to 35 years old (1), 36 to 55 years old (2) and over 55 years old (3) 125
Figure 6.1. Decision tree, predicted probabilities of voting for a candidate 137
16 Figures and Tables
Tables
Table 2.1. Results across different treatment conditions 60
Table 2.2. Percentage of respondents that express that they would vote for the corrupt candidates across the different type of questions 64
Table 2.3. Multinomial logistic regression for tradeoff argument 64
Table 3.1. Characteristics of the samples 72
Table 3.2. Randomization Test. Mlogit Regression Model. Dependent Variable: Corruption Treatment 73
Table 3.3. Attributes and text for each component 74
Table 3.4. Descriptive of Dependent Variable: vote probability. 77
Table 3.5. Predicted probabilities and the relative reduction of corruption 81
Table 3.6. Derivatives expressed as a semi-elasticity 82
Table 4.1. Wording of the experimental vignettes 95
Table 5.1. Summary of expectations according to each theory 120
Table 5.2. Distribution of respondents across gender, political sophistication and age groups 122
Table 5.3. Nested model comparison 125
Table 5.4. Model including triple interaction with gender 127
Table 5.5. Models including triple interaction with political sophistication and age 128
Table 6.1. Descriptive of Dependent Variable: choice 138
Table 6.2. Multinomial logit for choice when clean alternative is available 139
17
Chapter 1
Introduction
According to democratic theory, we would expect voters to hold dishonest politicians accountable because corruption is a clear signal of an underperforming government that will not act in the voters’ best interests. However, empirical evidence shows that voters around the world barely punish corrupt politicians at all at the polls. The aim of this thesis is to assess citizens’ attitudes towards corruption and its relative importance on their voting choices, in order to provide a better understanding of accountability regarding corruption.
Recurring accounts for why voters do not sanction corruption more severely posit that they are ignorant about wrongdoings and who to blame for them, or that they are actually not all that worried about malfeasance. Conversely, the argument of this dissertation is that, ceteris paribus, voters prefer the honest politician. However, the ceteris paribus condition, is seldom realistically fulfilled. In elections, voters do not face a situation where solely their position towards corruption determines their voting intention, but a multidimensional choice between many different potentially relevant factors.
The first section of this chapter (1.1 The Puzzle), justifies the importance of addressing this research question: the need to curb corruption and the consequent importance of better understanding corruption accountability in
19 Introduction
elections. In section 1.2, Existing explanations, I discuss some of the recurring interpretations of why voters do not sanction dishonest politicians more. I briefly review the arguments and some of the empirical evidence supporting them, and highlight some of their shortcomings. Section 1.3, Overview of the Argument, illustrates the argument of this dissertation, which will be tested in the following empirical chapters. Section 1.4, Contribution to the tradeoff argument lists up to six contributions of this dissertation to the state of the art. In section 1.5, Research Design, I justify the suitability of using survey experiments run in Spain to address my research question. Therefore, I justify both the case selection and the techniques used. Finally, section 1.6, Outline of the Dissertation presents an overview of the structure of the dissertation and a very brief summary of each of the empirical chapters that follow this introductory chapter.
1.1 The Puzzle
Corruption, defined as “the abuse of public office for private gain” (World Bank), has severe negative consequences for society. It hinders economic development (Mauro, 1995; Wei, 2000), increases economic inequality and poverty (Gupta et al., 2002), undermines workers’ welfare (Fisman & Golden, 2017) and citizens well-being in general (Tay et al., 2014). Corruption also erodes trust in government and other political institutions (Andersen & Tverdova, 2003; Ares & Hernández, 2017; Mishler & Rose, 2001), and negatively impacts the evaluation of the political system (Andersen & Tverdova, 2003). From a political philosophy perspective, the worst consequence of corruption is arguably that it ultimately obstructs the fundamental democratic principle of impartial access to political institutions and public services (Rothstein and Varraich 2017). This is because any involvement of a politician or a public server in a corrupt activity directly entails some citizens being favored over others.
In highly corrupt countries with widespread bureaucratic corruption, citizens are often expected to pay bribes to access basic public services to which they are legally entitled, such as healthcare or education (Fisman and Golden, 2017). This situation undoubtedly creates inequalities between those that can afford to pay the bribes and those that cannot. However, the indirect effects of political
20 Chapter 1
corruption can be even more damaging than the direct cost of bribes (Fisman and Golden, 2017). Corrupt activities at higher levels of the political structure do not involve citizens’ direct payments, but they do negatively affect citizens’ lives by favoring disreputable companies over more competitive options, building unsafe public infrastructures or wasting public money that could be invested elsewhere (Rose-Ackerman & Palifka, 2016). The collapsing of the Rana Plaza building in Bangladesh is perhaps one of the worst recent tragedies to illustrate the disastrous impact of corruption on citizens’ lives. The violation of all manner of regulations and laws in the construction and operation of this shockingly unsafe building led to its collapse in April 2013, killing 1,130 people and injuring other 2,500 (Fisman & Golden, 2017).
Despite its negative consequences for society and its direct negative impact on citizens’ lives, corruption is still widespread around the world and affects all types of political and economic systems (Shleifer & Vishny, 1993). According to TI’s global barometer, in 2013, 27% of respondents reported having paid a bribe in the last 12 months (Pring, 2017). This shows that corruption has had a direct impact on the life of at least one in four people in the 170 countries included in the analysis. To this, we should add all the people affected by grand corruption1. This is more difficult −or indeed impossible− to measure and is therefore not accounted for in this study.
The development of strategies to combat corruption has grown in importance (Kpundeh, 1998) and supranational institutions such as the World Bank, the OECD and international NGOs now include the fight against corruption among their primary objectives. Since the 90s, research on corruption has studied the consequences and causes of corruption and the remedies for it. An extensive literature currently addresses these questions and aims to grow further in order to combat corruption. Scholars and practitioners have certainly learned many
1 The literature differentiates between grand (or political corruption) and petty corruption (or bureaucratic corruption). Grand corruption refers to the activities that “take place at the high levels of the political system” (Amundsen 1999:3). These activities are carried out by politicians and state agents at the top level of the state. Conversely, petty corruption refers to the corrupt activities in the public administration implemented by civil servants and other public sector workers. A typical example is when citizens are asked to pay bribes in order to access certain public services or to fast track their petitions.
21 Introduction
things in the last two decades. Nevertheless, it is still unclear what the best mechanisms are to curb corruption, and there are cases where anticorruption policies have failed spectacularly (Persson et al., 2013).
One of the key questions that must be addressed further is the presence of corruption in democratic countries. A common expectation of political science is that democratic institutions should be curbing political malfeasance via electoral responsiveness and accountability. However, corruption is also widespread in democratic countries. Figure 1.1 shows the correlation between levels of democracy in a country and Transparency International’s Corruption Perception Index (CPI)2 (Teorell et al., 2013). Although there seems to be a certain relationship between levels of democracy and corruption, it is startling that a not negligible number of countries are extremely democratic but only have average levels of transparency (marked in green in Figure 1.1).
Even if one is skeptical about the miracles of democratization and is aware of the many challenges that this political system faces, from a purely theoretical point of view, it is puzzling how corrupt governments can survive in democratic societies, since one of the main functions of free elections, a fundamental institution in every democracy (Dahl, 1971), is to hold governments accountable.
2 The 2020 QOG Standard dataset was used to create this figure. Level of Democracy is an index that combines the democracy index from Freedom House and the one from Imputed Polity. The scale ranges from 0 to10, where 0 is least democratic and 10 most democratic. The scale of Transparency’s International Corruption Perception Index ranges from 0 to 100, where 0 equals the highest level of perceived corruption and 100 equals the lowest level of perceived corruption.
22 Chapter 1
Figure 1.1. Correlation between Level of Democracy and Corruption Perception
Index
10
8
6
4
LevelDemocracy of
2 0
0 20 40 60 80 100 Corruption Perception Index
The main principle of democratic systems is that citizens should be able to have a say in the development of the laws or policies in their countries (Giné & Mansuri, 2018). Because of this, a distinct characteristic of representative democracies is that governments are selected through free and universal elections (Manin et al., 2012). Advocates of liberal democracy consider elections to be a unique instrument to ensure citizens’ rights, but also a medium for obtaining good outcomes for society through selecting and sanctioning governments (Healy & Malhotra, 2013). These two functions of elections explain why governments should represent the interests of citizens. On one hand, electors should be using their vote to select the candidates that best represent their values and policy preferences or that have the best traits; on the other hand, electors should be sanctioning representatives that underperform. Furthermore, fear of not gaining reelection should induce those in power to enact voters’ preferences while in office (Manin et al., 2012).
23 Introduction
Whether we consider elections as essential tools to select representative or well- performing governments, or as tools to retrospectively sanction poor governments, it is theoretically perplexing why an informed and free citizen would vote for a corrupt incumbent: corruption is both a clear signal of a bad government that will not act in the voters’ best interests, and also a clear motive to sanction the government in power (Fearon, 1999). Yet, empirical evidence shows that voters around the world barely punish corrupt politicians at all at the polls (Bågenholm, 2013; Chang, Golden, & Hill, 2010; Costas, Solé-Ollé, & Sorribas-Navarro, 2009; Dimock & Jacobson, 1995; Eggers & Fisher, 2011; Golden, 2010; Peters & Welch, 1980; Reed, 1999; Rivero & Fernández-Vázquez, 2011). For example, different studies that have examined the effect of corruption scandals in the US at different moments in time have identified an electoral punishment of members of the House of Representatives of between 5% and 11% (Dimock & Jacobson, 1995; Peters & Welch, 1980). Politicians involved in the British parliamentary expense scandal only suffered a 1.5% vote share loss (Eggers & Fisher, 2011), and the cost of corruption allegations for Italian representatives from 1948 to 1994 was about 5% (Chang et al., 2010).
The rather lenient punishment for corruption at the polls could lead one to think that citizens do not care about malfeasance. Nevertheless, when interviewed in surveys, citizens are able to identify malfeasant activities, express their clear rejection of corruption and little intention of voting for corrupt politicians (Muñoz et al., 2016; Weitz-Shapiro & Winters, 2016; Winters & Weitz-Shapiro, 2013). As well as citizens rejecting corrupt activities in surveys, scholars also agree on the negative effects of corruption on individuals’ political attitudes. Corruption, and especially individuals’ perceptions of it, have been linked to negative attitudinal reactions such as lower trust in politicians, in political parties, and in representative institutions (Ares & Hernández, 2017; Mishler & Rose, 2001), to negative evaluations of the performance of the political system and government (Andersen & Tverdova, 2003), or dissatisfaction with the functioning of democracy (Villoria et al., 2013). In this vein, Pattie and Johnston (2012) find that the UK MPs’ expenses scandals not only affected the evaluation of the politicians involved, but also altered the respondents’ emotional reactions, as the whole affair made respondents very angry.
On one side, corruption is condemned and seen to have a negative impact on citizen’s attitudes. At the same time, this coexists with very lenient punishment
24 Chapter 1
of corrupt politicians in elections. This is clearly a puzzle for the social sciences (Kurer, 2001). My dissertation aims to further the understanding of why the effect of corruption on individuals’ attitudes does not have the behavioral implications that we might expect.
1.2 Existing Explanations
Existing attempts to understand the all too frequent reelection of corrupt politicians around the world focus on voters’ ignorance – their lack of information and inability to attribute responsibility – or on their indifference towards corruption. According to the latter position, voters simply do not care much about corruption. In the next two sections I review these arguments and some of the empirical evidence that supports them, and indicate some of their shortcomings. After that, I present an overview of the argument of this dissertation, that will be tested in the following empirical chapters.
1.2.1 The ignorant voter
To explain the paradox of corruption being only barely punished in elections, a strand of literature has focused on citizens’ ignorance of the actual corrupt acts committed by politicians or on their inability to attribute responsibility correctly. One of the clear limits of electoral accountability is the informational disadvantages that voters have with respect to elected officials (Ferejohn, 2012). These informational inequalities are even wider when it comes to corruption, because of the high incentives that politicians have to conceal their criminal activities. Clearly, we cannot expect citizens to hold politicians to account for malfeasant behavior if they do not know about that behavior. However, even when information about the politicians’ integrity is actually available, the story of the electoral punishment of corruption is not as straightforward as might be expected. In fact, in the extant literature, there is little consensus on voters’ reactions to information about corruption.
25 Introduction
Increased news coverage of corruption scandals has shown to decrease the vote share of local corrupt politicians (Costas-Pérez et al. 2012). A quasi-experiment in Brazilian municipalities, Ferraz and Finan (2008) indicates that the electoral consequences of corrupt practices are magnified in those municipalities where local radio divulged information about the audit. Similar results have been obtained with survey experiments: when respondents find out about corruption, they punish the dishonest politicians involved (Winters & Weitz-Shapiro, 2013). However, a set of field experiments that used fliers or newspaper advertisements to inform citizens about corruption did not find any increased electoral punishment (e.g. Boas, Hidalgo, and Melo 2018; Dunning et al. 2019). Actually, one study even found that informing the public about corruption reduced turnout and, as a consequence, decreased the vote shares for both the corrupt incumbent and the challenger (Chong et al., 2014).
These divergent results cannot be explained away by the different levels of credibility commanded by the information provided. Research has found that accusations have greater impact on support for corrupt politicians when the information comes from a trusted source (Botero et al., 2015; Weitz-Shapiro & Winters, 2016). While the credibility of the information seems to be essential, it is not enough to enhance accountability: in the field experiments discussed above, even information stemming from a credible source obtained no effects. Furthermore, the hypotheses based on the availability and credibility of information are incapable of explaining changes in individuals’ political attitudes. If corruption does have an effect on political attitudes, it must be because citizens are aware of corruption.
Trying to make sense of why these changes in attitudes do not translate into behavioral responses, one strand of literature focuses on what settings make it easier for citizens to apportion responsibility for corruption (e.g. Ferrer 2020; Schwindt-Bayer and Tavits 2016; Xezonakis, Kosmidis, and Dahlberg 2016). To be able to punish corruption at the polls, voters need to know who to blame for corruption. However, institutional settings provide differing opportunities and incentives to monitor and punish politicians (León & Orriols, 2019; Powell & Whitten, 1993). To study how institutional contexts favor or impede the attribution of responsibilities, a line of research has drawn on the notion of “clarity of responsibility”. Clarity of responsibility is a property that captures the extent to which citizens can apportion responsibility for politicians’ actions. Low
26 Chapter 1
clarity of responsibility allows politicians to shift the blame onto other actors for their actions. Different factors—such as the characteristics of the electoral and party systems, the vertical fragmentation of power or the number of veto players, among others—usually determine a given institutional system’s clarity of responsibility. According to this literature, when clarity of responsibility is high—when the government has been in power for a longer time and is controlled by a single majority—there is a stronger negative effect of generic corruption measures on the likelihood of voting for the incumbent; conversely, when clarity of responsibility is low—in proportional electoral systems and contexts of higher party fragmentation—the impact of corruption on the likelihood of voting for the incumbents is weaker (Schwindt-Bayer & Tavits, 2016; Tavits, 2007; Xezonakis et al., 2016). However, these studies usually focus on the aggregate perception of corruption in a country and not on concrete corruption scandals. The clarity of responsibility hypothesis is unable to explain why in field experiments, even when receiving fliers with clear and credible information about the incumbent’s specific wrongdoings, there is no effect on the electoral outcome.
In fact, the ignorant voter argument cannot account for why, even when information is available and voters are aware of the particular involvement of certain politicians in scandals, a substantial number of citizens tend to forgive corrupt politicians at election time. This was, for instance, the case in the UK parliamentary expenses scandal. Even though information was available and voters were aware of the implication of particular MPs in the scandal, they were only marginally less likely to support those MPs (Vivyan et al., 2012). This dissertation centers precisely on why even informed citizens condone corrupt candidates. Clearly more research is needed to understand why campaigns based on distributing information might fail to enhance electoral accountability. Furthermore, it seems more interesting to research factors that might explain why citizens would knowingly decide not to sanction corrupt politicians than to look for explanations based on voters’ ignorance of government performance.
27 Introduction
1.2.2 The indifferent voter
The surprisingly lenient punishment of corruption in elections, even when information is available, has extended the idea that voters do not care much about a politician’s integrity (Fisman & Golden, 2017). This is a recurring statement in the public debate. Headlines such as “Corruption allegations? Voters don’t seem to care” are common when election day is approaching (Seidman, 2018). While scholars tend to have a more contained message than journalists, the idea that voters do not care as much about corruption as we might expect is also an underlying assumption in some academic work. This position becomes especially evident when researchers express worry about the efficacy of surveys in collecting truthful answers due to social desirability (e.g. Boas, Hidalgo, & Melo, 2018; Incerti, 2019). According to this position, citizens’ clear opposition to corruption in survey is a result of respondents will to provide socially accepted answers rather than a reflection of their truthful opinions.
When interviewed in surveys, citizens express their clear rejection of corruption and their low possibility of voting for corrupt politicians (Muñoz et al., 2016; Weitz-Shapiro & Winters, 2016; Winters & Weitz-Shapiro, 2013). Results of round 3 of the Afrobarometer are a great example of the ability of citizens to identify corrupt activities and their propensity to disapprove of those activities. The survey, run across 18 African countries in 2005, shows that a clear majority of interviewees across all countries considered different corrupt activities as wrong and punishable (see Figure 1.2 to 1.4). 75 percent considered it wrong and punishable if a government official gave a job to an unqualified family member. 76 percent of respondents also stated it was wrong and punishable if a government official demanded a favor or an additional payment for some service that was part of her job. And up to 60 percent of respondents considered it wrong if an official decided to locate a development project in an area where her friends and supporters lived. There are of course variations across countries, however the patterns show that the majority of respondents reject corrupt activities and want officials to be punished for them (Afrobarometer Round 3). Similar results were also obtained in the World Values Survey Wave 6, where a clear majority of respondents across all counties considered that it was never justifiable for “someone [to] accept a bribe in the course of their duties” (see Figure 1.5) (World Values Survey Wave, 6, V202).
28 Chapter 1
Figure 1.2. Percentage of respondents that consider that if “a government official gives a job to someone from his family who does not have adequate qualifications”, this is wrong and punishable, wrong but
understandable, or not wrong at all
80
60
40
Percent of respondentsPercent of
20 0 Wrong and punishable Wrong but understandable Not wrong at all
Note: Own elaboration with data of Afrobarometer Round 3.
29 Introduction
Figure 1.3. Percentage of respondents that consider that if “a government official demands a favor or an additional payment for some service that is part of his job”, this is wrong and punishable, wrong but
understandable, or not wrong at all
80
60
40
Percent of respondentsPercent of
20 0 Wrong and punishable Wrong but understandable Not wrong at all
Note: Own elaboration with data of Afrobarometer Round 3.
30 Chapter 1
Figure 1.4. Percentage of respondents that consider that if “a public official decides to locate a development project in an area where his friends and supporters live”, this is wrong and punishable, wrong but
understandable, or not wrong at all
80
60
40
Percent of respondentsPercent of
20 0 Wrong and punishable Wrong but understandable Not wrong at all
Note: Own elaboration with data of Afrobarometer Round 3.
31 Introduction
Figure 1.5. Percentage of respondents that consider that it is never justifiable to
accept a bribe. (1: it is never justifiable and 10: it is always justifiable)
80
60
40
Percent of respondentsPercent of
20 0 Never justifiable 3 4 5 6 7 8 Always justifiable
Note: Own elaboration with data from World Value Survey Wave 6, V202.
A recurring interpretation of the gap between the overwhelming disapproval of corruption in surveys and the lenient punishment for it in elections is that surveys are plagued with social desirability bias (see for example: Chong et al. 2014; Boas, Hidalgo, and Melo 2018; Incerti 2019). According to this position, the concern that citizens express in surveys is driven by a will to express socially accepted attitudes. However, in elections, where citizens can conceal their individual behavior, they vote for corrupt politicians. An underlying assumption of this interpretation is that citizens do not care about the integrity of governments and that they fully intend to vote for a corrupt politician.
While the idea of an indifferent voter would easily explain the gap between what is reported in surveys and what is observed in elections, I argue that the story behind this puzzle is somewhat more complex. The question of why voters do not punish corruption is based on the understanding of elections as an accountability mechanism, one where citizens take a politician’s past performance into consideration in order to determine whether they reelect the
32 Chapter 1
incumbent. According to the reward-punishment model, voters reelect high- performing politicians and punish unsuitable candidates by ejecting them from office. As corrupt behavior is an unequivocal signal of an underperforming government (Fearon, 1999), we would expect concerned voters to punish dishonest incumbents. However, this dissertation demonstrates empirically that most citizens do care about politicians’ integrity and that ideally they would not vote for a corrupt politician. But, because integrity is only one of the many bases for evaluating a government’s performance, under some circumstances, even voters that reject corruption can reelect a corrupt politician. Therefore, this dissertation rejects the idea that voters that care about corruption should punish corrupt governments no matter what. This assumption expects voters to act as though corruption were the only issue that should be considered when casting a vote. However, elections are a multidimensional phenomenon where voters have to consider many more aspects than only a candidate’s integrity. The argument of this dissertation is discussed in more detail in the next section.
1.3 Overview of the argument: the multidimensional voter
In representative democracies, citizens delegate the task of governing to elected public officials. These officials are put in charge of deciding and enforcing the rules that govern everyone. Elected representatives have the power to make decisions about almost every single aspect of citizens’ lives: how their children are educated, what medical service they are provided with, how they can move from one place to another etc. (Manin et al., 2012). But voters have only one chance to express their opinions. They have to use the vote to both reward or punish the incumbents’ performance and to select the option they consider best (Fearon, 1999). As such, a voting decision is a multidimensional phenomenon, in which citizens have to take many different dimensions into account (Hainmueller et al., 2014).
As it is impossible to consider all the dimensions of a government’s performance, voters limit the number of aspects that they devote attention to (Singer, 2011). The vote is a function of what individuals care about, “the intensity of this concern”, and the candidates’ positions regarding these concerns
33 Introduction
(Rundquist, Strom, & Peters, 1977:956). Citizens evaluate politicians by weighing up their positions and then voting for the candidate that best represents their preferences both in terms of traits and policies. As integrity is not the only basis for evaluating a government’s performance, under some circumstances, even voters that reject corruption can reelect a corrupt politician.
The argument of this dissertation is that citizens do indeed care about malfeasance and would ideally never have to vote for corrupt politicians. However, they also have other concerns. While integrity is a clearly valued aspect of government, other aspects such as a politician’s ideological position, their performance managing the economy, international relations or internal conflicts are also important factors to consider. In elections, when casting a vote for one political option, citizens face tradeoffs, as they only have one chance to express their choice but many different preferences to cover. As a result of this situation, voters may decide to overlook a candidate’s wrongdoings if she ranks very well in terms of other important preferences. The literature has branded this as an implicit trading off of a candidate’s integrity against other valued characteristics (Rundquist et al., 1977). As such, the lenient punishment of corrupt politicians in elections should not be attributed to citizens’ indifference to integrity, but rather to the tradeoffs they face in elections (see Chapter 3 for a more detailed discussion). Furthermore, as demonstrated in Chapter 2, the gap between the rejection of corruption in surveys and citizens’ behavior in elections is not due to citizens concealing their true attitudes, but due to the inability of standard survey questions to reflect the tradeoffs that citizens face in elections: they do not depict the relative importance of corruption.
The conception of elections as a multidimensional phenomenon is not novel to this dissertation. In fact, the first discussion of the relative importance of corruption on the vote dates from as early as the 1970s (Rundquist, Strom, and Peters, 1977). However, the literature on accountability for corruption has mostly ignored this approach and it has not been until recently that scholars are now once more incorporating it into their analyses. Studies that take this approach show that citizens trade off integrity in good economic contexts (Klašnja & Tucker, 2013), when the politician has a good management record (Muñoz et al., 2016), or when she implements favorable economic policies (Konstantinidis & Xezonakis, 2013). Nevertheless, research linking the tradeoffs that citizens face in elections and the lenient punishment of corrupt politicians
34 Chapter 1
is still quite scarce and many questions have yet to be addressed. For example: what are voters trading integrity against? What exactly does the tradeoff argument entail? What individual conditions increase the tradeoff? And what characteristics of the alternative candidates increase it? These questions are empirically addressed in this dissertation, providing a clear contribution to the state of the art.
However, the argument of this dissertation is not only related to the multidimensionality of elections, but also to the limitation of elections as a mechanism for keeping corruption in check. Most of the research that assesses why corrupt politicians are reelected seems to assume that elections are a mere accountability mechanism, where voters reward and punish politicians for their past performance. While this might be one way to use elections, voters can also use elections to select the politicians they think will best perform in office. In other words, voters might also use elections to select “good types” of politicians rather than to hold them directly accountable for their past performance (Fearon, 1999). Even if the vote were mostly used to hold politicians to account for their past performance, as discussed before, voters have to consider politicians’ performance in many other aspects than just corruption. The multidimensionality of elections clearly undermines elections as an accountability mechanism for corruption. Societies that aim to curb corruption should therefore implement horizontal accountability mechanisms, such as a strong and independent anti-corruption agency. This argument is discussed in more detail in the conclusions of this dissertation (Chapter 7).
35 Introduction
1.4 Contribution to the tradeoff argument
The first contribution of this dissertation is that it argues and provides empirical evidence that the gap between what observed in surveys and in elections is not explained due to a social desirability bias. But it can in fact be explained by the inability of standard survey questions to reflect the tradeoffs that citizens face in real elections and their consequent inability to measure the relative importance of corruption on the vote. Respondents provide truthful answers when expressing their rejection of corruption in surveys. However, in actual elections, corruption is just one of the many aspects they have to consider when casting a vote. Therefore, the answer to the gap between attitudes towards corruption and electoral behavior must be found in the possible tradeoffs voters may face in elections.
While previous literature has already discussed the possibility of voters trading corruption for other valued characteristics, the literature has so far not agreed on exactly what the tradeoff hypothesis entails. Some studies focus on the factors that moderate punishment of corrupt politicians at the polls (e.g. Zechmeister and Zizumbo-Colunga, 2013), while others discuss the relative importance of corruption (e.g. Fisman and Golden, 2017). However, these two ways of understanding the tradeoff hypothesis entail different causal mechanisms and may, therefore, entail different consequences for policies that aim to combat corruption by increasing citizens’ awareness. The second contribution of this dissertation is that it digs deeper into the comprehension of the tradeoff hypothesis by disentangling these two ways of understating this theory and by discussing their implications.
The relative weight argument proposes that corruption is just another dimension that voters consider when casting their votes, and that they might give more importance to issues other than corruption (e.g. Fisman and Golden, 2017). This explanation draws from the argument of Rundquist et al. (1977), which assumes a rational voter that weighs up the candidates’ characteristics and votes accordingly. According to this argument, voters choose a corrupt politician when they place more weight on characteristics other than her integrity. The second approach, which I call the “conditional punishment argument”, proposes that other positive characteristics may diminish the negative effect of corruption
36 Chapter 1
on the support for a candidate. This argument is in keeping with the research, showing that a good economic situation conditions the negative evaluation of a corrupt politician (e.g. Zechmeister and Zizumbo-Colunga, 2013). In this case, the causal mechanism is less rational, as voters can vary how they weigh up a candidate’s alleged corruption depending on the candidate's other qualities. Both arguments have different substantive implications. The relative weight approach conceives a rational voter that simply chooses to overlook corruption, while the conditional punishment argument conceives a voter that might be unconsciously driven by psychological bias, such as the will to avoid cognitive dissonance. The different implications for anti-corruption campaigns based on these two conceptions are discussed in Chapter 3.
This dissertation’s third contribution to the tradeoff argument is to identify the factors against which voters are trading integrity. Early work by Rundquist et al. (1977) discussed an implicit exchange between the integrity of the candidate and her position on certain policies or issues. Subsequent literature focused on a tradeoff between integrity and the competence or the economic performance of the candidate (e.g. Rosas and Manzetti 2015; Muñoz, Anduiza, and Gallego 2016; Konstantinidis and Xezonakis 2013). These studies clearly follow the popular expression of “rouba, mas faz” (he steals but he delivers) but neglect other type of tradeoffs that could be at play in elections. In addition, although previous studies conceive elections as a multidimensional phenomenon, they have usually only focused on one tradeoff at a time. In reality, voters are likely to face different tradeoffs all at once in an election. They may value the management experience and policy position of one candidate, and the integrity of another. As they can only vote for one politician, they will have to trade off some of the qualities that they value. This dissertation assesses different tradeoffs in one multidimensional experimental setting. I argue that voters are likely to trade off integrity against both partisan preferences and good economic performance. Both co-partisanship and good economic performance clearly determine voting choices and to some extent condition the negative effect of corruption on support for a politician.
The fourth contribution of this dissertation is to extend the “he steals but he delivers” argument. Previous literature has usually focused on assessing whether citizens condone corruption when the malfeasant politician manages to deliver desirable public goods and outcomes such as economic growth (e.g: Klašnja &
37 Introduction
Tucker, 2013; Muñoz, Anduiza, & Gallego, 2016; Zechmeister & Zizumbo- Colunga, 2013). I extend this line of research by analyzing citizens’ propensity to vote for politicians that manage to provide desirable outcomes precisely due to their wrongdoings (“he delivers because he steals”). While political malfeasance has clear negative consequences in the long run, in certain situations breaking the law can be perceived as actually leading to a superior societal outcome in the short run3. Nevertheless, most of the literature has ignored this fact (cf. Fernández-Vázquez, Barberá, & Rivero, 2016). Furthermore, I argue that changes in character evaluations of politicians are a relevant mechanism for explaining the impact of malfeasance on voting choice. A decrease in trustworthiness is likely to drive a decrease in the probability of voting for a malfeasant politician.
The fifth contribution to the tradeoff argument made in this dissertation is a focus on the characteristics of the alternative candidates. Most of the literature so far has ignored the importance of the qualities of the alternative candidates. Since punishing a corrupt politician either requires voters to vote for an alternative candidate or to abstain, I argue that the attributes of alternatives candidates influence voters’ reactions to malfeasance. Whether or not a clean and attractive alternative is available, in terms of party preferences and economic performance, determines whether voters punish the corrupt politician or not.
The sixth and final contribution to the tradeoff argument is to draw an even more precise picture of the relative importance of corruption across different individual characteristics. The dissertation identifies some modest but potentially relevant heterogeneities in citizens’ responses to corruption. It also shows that the findings – that voters care about corruption but in multidimensional decisions they are willing to trade off integrity – are in fact universal among different types of people.
All in all, this dissertation shows that if we embrace a simplistic view of elections, we are expecting voters who care about corruption to use the one vote they have to hold the government accountable. By adopting a multidimensional
3 First, most of the negative consequences of malfeasance take some years to materialize or to be perceptible. Second, because politicians can justify their malfeasant behavior by claiming it to be in the best interests of the community.
38 Chapter 1
perspective of elections, this study shows that under certain circumstances, voters can reject corruption and at the same time reelect a corrupt politician.
1.5 Research design
1.5.1 The Spanish case
Why Spain?
The original data collected for this dissertation are from Spain. However, this dissertation is not about Spain. The data are gathered in this country because, as discussed below, it is an ideal case-study to examine why voters in free elections vote for corrupt politicians. As such it is the best setting to test my research question. I decided to concentrate on only one context, since this is the most adequate design to test causal effects. The scope of this dissertation is however much broader.
The reach of this dissertation is individuals that can vote in democratic countries with free elections and medium to low levels of corruption. As the query that drives this research is to understand why citizens vote for corrupt politicians, there is no need to justify why the scope is limited to countries that hold free elections. Additionally, the focus is on citizens that live in contexts with medium to low levels of corruption, because it is in these settings that this query is especially compelling. While corruption scandals can occur in these settings, political malfeasance is not too excessively widespread; as such, in elections, there is always a clean alternative to vote for. This is a crucial point since the explanation for why citizens choose corrupt politicians would be completely different in settings where all politicians are facing accusations of corruption (Pavão, 2015). Accordingly, although this dissertation mainly uses data from Spain, its results could be generalized to other countries with medium levels of
39 Introduction
corruption and free elections4. Nevertheless, one must keep in mind the specificities of each context that can alter the relevance of the different tradeoffs voters may face in elections.
For several reasons, I consider Spain a good case to study why voters in free elections vote for corrupt politicians. Spain is a democratic country that regularly holds free election and that, at the same time, presents medium levels of corruption. It scores 58 on a scale of 0 (highly corrupt) to 100 (very clean) according to Transparency International’s 2018 Corruption Perception Index (see Figure 1.6 for a comparison of CPIs across countries). Corruption scandals are often a salient issue in Spanish politics (see e.g. Ares and Hernández, 2017), and they were at the top of the agenda at the time the data for this dissertation were gathered. This makes the topic relevant and realistic, as Spanish citizens are faced with these issues when casting their vote.
Furthermore, Spain represents a typical case of a place where citizens report a highly negative view of corruption in surveys but often do not hold malfeasant politicians accountable in elections, hence making it an appropriate setting for the research question of this dissertation. According to the data of the official survey institute of the Spanish government (CIS), for years, corruption has been among the three most important concerns of Spanish citizens. Up to 50% of the respondents of the CIS Barometer considered it the most important problem in Spain in March 2015 (see Figure 1.7 for the evolution of the percentage of respondents that consider corruption the main problem in Spain). Similarly, in a 2011 survey conducted by the same institute (CIS 2905), 87% of the respondents considered corruption a problem of paramount importance in Spanish democracy. According to the same survey, only 10% of the respondents would vote for a corrupt but efficient candidate rather than an honest but inefficient candidate.
In actual election results, however, punishment at the polls is very limited if at all present (Costas-Pérez et al., 2012; Rivero & Fernández-Vázquez, 2011). According to Costas-Pérez et al. (2012), the actual punishment of corrupt
4 Corruption Perception Index (CPI) between approximately 80 and 45. The CPI is an index developed by Transparency International that ranges from 0 to 100, where zero means highly corrupt and 100 means very clean.
40 Chapter 1
mayors in Spain in elections is between 2.8 and 3.8%. Analyzing mayors under criminal investigation in two Spanish regions, Rivero and Fernández-Vázquez (2011) did not find any significant punishment in elections.
Finally, some excellent studies have already been conducted in Spain. The empirical chapters of this dissertation test hypotheses that have also been addressed in studies such as Anduiza, Gallego, and Muñoz 2013; Muñoz, Anduiza, and Gallego 2016 and Fernández-Vázquez, Barberá, and Rivero 2016. Conducting the research for this dissertation in the same country and keeping, to some extent, the institutional context constant facilitates the comparison of the results that I obtain. Being able to compare the evidence gathered in this dissertation is of paramount importance, as scientific knowledge only increases when we are able to reproduce similar results with different measurements (Sniderman, 2018).
Figure 1.6. Corruption Perception Index 2018
Note: Own elaboration with data of Corruption Perception Index 2018, TI.
41 Introduction
Figure 1.7. Percentage of respondents that consider that corruption is the most important problem Spain is facing
50
40
30
20
Percentage of respondents
10 0 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020
Note: Own elaboration with data of CIS Barometer, for March of each year.
Corruption in Spain
Political corruption in Spain is mainly grand corruption, i.e. corruption carried out by politicians and state agents at high levels of political institutions. This is clearly reflected in the low levels of victimization among the Spanish population. In countries with high levels of petty corruption − the type of corruption implemented by civil servants and other workers of the public sector − the number of citizens that paid a bribe in the last twelve months to a public official is usually quite high (Villoria & Jiméne, 2012). According to the data from TI's 2017 Global Corruption Barometer, Spain has very low (practically inexistent) levels of victimization (see Figure 1.8 for a comparison of Spain to other European countries). In comparison with other European countries that were included in the survey, Spain is among the countries with the lowest levels of victimization. Other countries such as Hungary, Romania and Lithuania, and
42 Chapter 1
regions such as Sub-Saharan Africa, Latin America and the Caribbean have much higher levels of victimization.
Nevertheless, when disentangling the Spanish data and assessing victimization in different public sectors, it is remarkable that up to 6 percent of respondents said that they had paid a bribe for a construction permit or in the property sector, while zero percent had paid a bribe to a police officer (Villoria & Jiméne, 2012). These results are a clear reflection of the fact that corruption in Spain is mainly related to the local real estate sector. Of the cases of corruption that were uncovered from 2004 to 2010, the most prevalent type of corruption was corruption related to urban planning (Villoria and Jiménez 2012). Furthermore, the decrease in the CPI score of Spain from 7.1 in 2004 to 6.5 in 2008 is, according to the TI’s annual report, mainly due to the increase in urban planning corruption (Jiménez, 2009). This does not mean that urban planning is the only source of corruption in Spain; other policy areas such as party funding and public procurement have recently emerged as especially prone to corruption scandals.
43 Introduction
Figure 1.8. Percentage of respondents (households in European countries) that paid a bribe in 2017 across different countries and regions
Russia Albania Ukraine Moldova Lithuania Bosnia and Herz. Armenia Romania Kazakhstan Hungary Serbia Belarus Azerbaijan Latvia Bulgaria Montenegro Slovak Rep. Kosovo Greece FYR Macedonia Czech Rep. Croatia Poland Italy Estonia Georgia Spain Slovenia Portugal Germany Cyprus
0 10 20 30 Total bribe rate
Note: Own elaboration with data from the 2017 Global Corruption Barometer, TI.
1.5.2 Data and methods
This dissertation mainly draws on experimental methods, and specifically on survey experiments, because this methodology offers advantages that are essential for addressing the research queries of this study. Experiments allow causal research questions to be tackled that cannot be addressed with observational data, and are particularly suitable for studying the underpinnings of individual political behavior (Morton & Williams, 2010). Furthermore, in this dissertation, experiments that are embedded in population-based surveys are favored over other experimental techniques, such as laboratory or field experiments, due to their ability to generalize to a target population (Mutz, 2011). In addition, because of their large and varied samples, survey experiments are
44 Chapter 1
also ideal for testing different causal explanations and thus for unpacking the causal mechanisms at play (as done in Chapter 4), identify moderating relationships (as done in Chapter 3) and subgroup differences (as done in Chapter 5). In this section, I develop each of these points and justify the use of survey experiments in detail.
The primary justification of the use of experiments is that the main research query of this dissertation is an inherently causal question, as it aims to identify the relative importance of corruption on voting choice. With observational data, it would be much more difficult to test this causal relationship. Observing that certain wrongdoings are correlated with a decrease in the vote share of the politicians that committed them is not the same as establishing that those wrongdoings influence voters’ likelihood of voting for those politicians. Corrupt politicians may differ in many ways from honest politicians and it could be that these differences are responsible for the decrease (or not) in their vote share. For example, corrupt politicians may promote more economic activity due to their willingness to defy established legal procedures. Voters may keep voting for these politicians not because of their actual wrongdoings but because of the side benefits of those wrongdoings (Chapter 4 or Fernández-Vázquez, Barberá, and Rivero 2016). Another possibility could be that corrupt politicians receive fewer transfers from the central government. Therefore, the reduction in the vote share of these politicians could be explained because of their reduced budgets rather than to the actual loss of reputation due to the wrongdoings themselves (Brollo, 2012). Only with experiments can we be completely certain how X affects Y: in this case how corruption affects the probability of voting for a politician (Morton & Williams, 2010).
Apart from the advantage of being able to address causal questions, experiments have been proven to be particularly suitable for studying political behavior, information processing and individual decision-making (Morton & Williams, 2010). These are exactly the topics this dissertation aims to explore. Experiments provide techniques that allow measuring, or at least come closer to measuring, individuals’ real preferences or behaviors, rather than eliciting their preferences in the abstract (McDermott, 2002). As experiments allow us to study certain topics without referring to them explicitly, it is possible to measure attitudes and behaviors without individuals rationalizing them.
45 Introduction
Particularly sensitive issues, such as corruption, that involve an ethical component and strong social norms, are extremely difficult to measure with standard survey questions. However, experiments have tools that allow researchers to come closer to truthful opinions, attitudes and behaviors, without respondents needing to rationalize their answers. My argument is that corruption does indeed entail strong social norms which have an impact on citizens’ preferences towards corrupt politicians (Boas et al., 2018). Nevertheless, in contrast to recurring interpretations, respondents are not afraid to disclose their true position to the researcher. They do actually care about corruption, and they have strong aversions to corrupt politicians, but certain situations make them trade their internal values regarding corruption against other qualities. The survey experiments that I use in the following empirical chapters are ideal to come closer to the decision-making processes in real elections, which would be extremely complex to measure with standard survey questions. In sum, it is essential for this dissertation to use survey experiments − such as the list experiment (in Chapter 2), the conjoint experiment (in Chapter 3, 5 and 6) and the dilemma experiment (in Chapter 4) − due to the unique techniques that they use to come closer to respondents’ real behavior.
Furthermore, one of the main methodological challenges of observational data and the study of the political consequences of corruption in general, is related to how corruption can be measured. Corruption is undesirable, unethical, and, in many cases, illegal. Therefore, the actors involved have great incentives to conceal it. This makes measuring corruption an extremely complex task. Objective measures, such as the number of corruption allegations being prosecuted, can be greatly biased by the willingness or capacity of the authorities to fight it. Countries with widespread levels of malfeasance are also likely to prosecute corruption less, either because their judicial system is also highly corrupt or because they do not have the capacity to handle all the various cases of corruption. As such, objective measures would underreport corruption in countries where the judicial systems do not put enough effort into controlling malfeasance and overreport it in countries with independent judicial systems that keep corruption in check.
On the other side, the frequently used subjective measures based on experts’ or survey respondents’ perceptions of corrupt practices, such Transparency International’s and the World Bank’s perceived corruption indexes, have been
46 Chapter 1
criticized for various reasons. First, these data do not measure actual levels of corruption, but citizens’ perceptions of corruption. Perceptions can of course be biased by citizens’ political preferences, their cynicism or their exposure to media reports (Treisman, 2007), creating artificial differences across countries or across time. For example, if perceptions are influenced by exposure to media reports, we could encounter the same problem as with the objective measures, i.e., the places where the media makes more effort to expose corruption could be incorrectly identified as being more corrupt. Furthermore, survey questions that ask respondents about their experiences with corruption have similar problems to the perception indexes. In addition, they could also be biased due to memory inaccuracy and by the fear of the consequences of exposing corruption (Treisman, 2007). Another drawback of measuring citizens’ experience with corruption for this dissertation is that these measures mainly capture corrupt activities that involve citizens, such as paying a bribe to access a public service. But they are ineffective for measuring wrongdoings that do not involve ordinary citizens, such as corruption at the political level, most common in countries with medium to low levels of corruption.
I choose to use survey experiments rather than laboratory or field experiments for a variety of reasons. While it is true that causal relationships could also be tested by using other types of experiments, survey experiments are particularly suitable for studies that aim to combine “the internal validity of experiments with the external validity of representative population samples” (Mutz 2011:5). By embedding experiments within representative surveys, it is possible to generalize beyond a narrow pool of subjects and represent a target population successfully. Pursuing representativeness would be extremely costly, or simply unfeasible, with laboratory experiments. Furthermore, using survey sampling techniques, survey experiments reach larger and more heterogenous groups of respondents than lab experiments. It is very difficult to reach diverse samples when using laboratory experiments. Conversely, the larger and more diverse samples used in survey experiments allow identifying even subtle differences among experimental groups, moderating relationships and individual heterogeneities, which are crucial goals of this dissertation.
In addition, survey experiments are an ideal tool for testing different causal mechanisms in one experimental setting given that, thanks to their strong statistical power, they allow a higher number of experimental conditions to be
47 Introduction
manipulated. This allows us to unpack the causal mechanism at play without having to sacrifice the representativeness of the sample. Testing for different causal mechanisms would be far more difficult to achieve in laboratory experiments due to their lower statistical power, and to do so with field experiments would be extremely costly.
Survey experiments also allow us to come closer to the treatments and the distractions of the real world than lab experiments, thus increasing the external validity of the experiment (Mutz, 2011). Nevertheless, just as with any methodological technique, survey experiments are not free of challenges (for a review on the advantages and limitations of survey experiments, see Sniderman, 2018). The main inconvenience of survey experiments, which also applies to lab experiments, is that they are based on hypothetical scenarios and therefore have problems in completely replicating behavior in real scenarios. The hypothetical bias arises in stated preference surveys when respondents behave in a different way in surveys (or in the lab) than they would in real life, because they are not faced with the actual cost that this behavior would have in a real situation (Incerti, 2019). For this dissertation, this would mean that respondents report greater willingness to punish corrupt politician than in reality. In each of the survey experiments used in this dissertation, I implement specific techniques to increase the cost for respondents to choose to punish the corrupt politician. In some experiments I try to reduce the hypothetical scenario by using actual party labels and common cases of corruption; in others I increase the strength of the tradeoffs that respondents are facing to better replicate the tradeoff they face in real elections. The external validity of this dissertation is also achieved by the consistency of the results obtained with experiments using different treatments, measures and samples. Nevertheless, I should highlight that, as discussed in the conclusions, the aim of this dissertation is not to identify the actual electoral cost that corruption has for politicians, but to assess whether citizens do care about corruption and if they would be willing to trade off corruption against other attributes that they value. As such, the actual estimates identified in this dissertation should be treated with some caution.
Some might argue that field experiments can overcome the hypothetical bias by increasing the external validity of the experiment (e.g. Boas, Hidalgo, and Melo 2018). However, these techniques have other drawbacks that would hinder the goals of this dissertation. The main disadvantage of field experiments is that they
48 Chapter 1
are extremely challenging to implement (Gerber & Green, 2012). They not only require significant economic resources but also the means to carefully monitor each step of the implementation process in the field. This is often impossible, and can lead to serious identification problems caused by respondents’ non- compliance or attrition. Non-compliance refers to subjects who do not take the conditions they are assigned to, i.e. they are treated even though they had actually been assigned to the control condition or vice versa. Non-compliance obviously creates serious problems when estimating the average treatment effect, especially in those situations where it is impossible to identify which subjects did not comply with the condition they were assigned to. Attrition occurs when the outcome measurements are missing for certain subjects. A problem is created “when attrition is systematically related to potential outcomes”, as in this case the remaining data is no longer a random sample of the original selection of subjects (Gerber and Green 2012: 211). Attrition is a recurring problem especially in field experiments because in these types of studies the outcome is measured over an extended period of time, thus increasing the respondents’ probability of dropping out of the study.
Furthermore, the high costs of implementing field experiments forces experimenters to keep the experimental conditions simple. A clear advantage of survey experiments over field experiments is that they make it easier to administer numerous variations of a treatment and as such, permit testing fine- grained theoretical propositions (Gerber & Green, 2012). Another important drawback of field experiments for this dissertation is their inability to manipulate certain issues in the real world. As field experiments measure real world outcomes, ethical concerns prevent experimenters using simulated information. While with survey experiments, I can easily assign random traits, corruption and performance information to candidates of different political parties, in field interventions I would be more limited in manipulating certain conditions.
In each of the empirical studies presented in this dissertation I use different strategies to increase their validity. The specific techniques used and the mechanisms employed to enhance internal and external validity of the results are discussed in the corresponding chapters.
49 Introduction
1.6 Outline of the Dissertation
This thesis is divided into seven chapters, including this introductory chapter, five empirical chapters where I discuss and test different aspects of my argument for the lenient electoral punishment of corruption, and the conclusions. The remainder of the thesis is organized as follows.
Chapter 2 assesses whether social desirability bias is behind the strong rejection of corruption in surveys. It does so by implementing two empirical strategies. First, a question is formulated in an unobtrusive way to increase respondents’ willingness to report their opinions truthfully. Second, a question is formulated as a tradeoff to mimic decision making in real elections. This chapter provides evidence that the gap between what is observed in surveys and in elections is due to the difficulties of standard survey questions to replicate the complexity of decision making in real elections.
Chapter 3 discusses and tests two possible tradeoffs that voters face in elections. By using the data of an original conjoint experiment, this chapter shows how respondents make decisions in a multidimensional scenario where they have to take several tradeoffs into consideration. This increases the external validity of the study, as it provides a more accurate account of what is happening in real elections where voters are confronted with multiple tradeoffs when casting their votes. This chapter demonstrates the relative importance of corruption in determining voting choice, and provides evidence for two tradeoffs that voters can face in elections.
Chapter 4 assesses respondents’ preferences when confronted with a political decision that reflects a tradeoff in the form of a dilemma between the integrity of a procedure and the societal outcome. This chapter also analyses the causal mechanisms that explain the link between knowing that a politician of the preferred party has defied the law and the likelihood of voting for her. This chapter provides evidence that changes in character evaluations of politicians are a relevant mechanism for explaining the impact of malfeasance on voting choice.
In Chapter 5 and 6 I evaluate the factors that decrease the relative importance of corruption on the vote. Chapter 5 assesses the correlation between citizens’ individual attributes and the relative importance they give to corruption when
50 Chapter 1
casting a vote, while Chapter 6 assesses the characteristics of the alternatives that increase or decrease the punishment of corruption at the elections. This chapter also disentangles the two forms of electoral punishment: voting for an alternative candidate or abstaining.
Finally, in Chapter 7, I review the findings and discuss the main contributions of this dissertation for the theories of electoral accountability.
51
Chapter 2
Social desirability bias
2.1 Introduction 5
According to a large number of empirical analyses, voters do not electorally punish corrupt politicians as much as expected by democratic theory (Chang, Golden, & Hill, 2010; Costas-Pérez, Solé-Ollé, & Sorribas-Navarro, 2012; Dimock & Jacobson, 1995; Eggers & Fisher, 2011; Golden, 2010; Peters & Welch, 1980; Reed, 1999; Rivero & Fernández-Vázquez, 2011). This situation could reflect citizen tolerance towards corruption, but surveys show that a large majority of citizens hold a negative view of corruption and, when asked, overwhelmingly report an intention to punish malfeasant incumbents in elections.
A possible explanation for this paradox may be social desirability bias, where interviewees prefer to report socially accepted attitudes (rejection of corruption) instead of truthful responses (intention to vote for corrupt politicians). However, an alternative argument suggests that standard survey questions are
5 Parts of this chapter are part of a work in progress: “They really care, among other things: Assessing social desirability bias in condemning political corruption” (written together with Eva Anduiza and Jordi Muñoz)
52 Chapter 2
simply not able to reflect the tradeoff citizens face in real elections. According to this argument, respondents are providing a truthful answer when expressing rejection of corruption in surveys. However, in actual elections corruption is just one of the many aspects they have to consider when casting a vote: i.e. survey questions are not able to estimate the relative importance of corruption.
This chapter assesses whether vote intention towards corrupt candidates, suffers from social desirability bias or whether standard survey questions are not successful in reflecting real elections. For this purpose, I use two empirical strategies. First, a list experiment that allows interviewees to be questioned in an unobtrusive way, removing the possible effects of social desirability. Second, I compare the effect of a corruption voting question framed as a tradeoff to a question that does not reproduce this tradeoff. The results show that when using a simple question (without the tradeoff) the expressed support for corrupt politicians is significantly smaller than when asked with a question that describes the tradeoff. Furthermore, when using questions framed as a tradeoff, the expressed willingness to vote for a corrupt candidate of the preferred party increases significantly no matter whether asked explicitly or in an unobtrusive way.
My reading of these results is that when people express negative views towards corrupt politicians they truly mean it and are not merely giving a socially desirable answer. In other words, they prefer, ceteris paribus, the non-corrupt candidate. When choosing a candidate in a specific election, however, other considerations play an important role in voter’s deliberations and the ceteris paribus condition is seldom realistically fulfilled. Therefore, the answer to the gap between attitudes towards corruption and electoral behavior must be found in the possible tradeoffs voters may face in elections.
53 Social desirability bias
2.2 Social desirability in attitudes towards corrupt politicians
Evoking truthful answers can be challenging when addressing sensitive issues, such as racism, homophobia, or corruption, as citizens may adapt their answers to social expectations or simply avoid answering these questions (Corstange, 2009). The effect of social desirability bias has been widely demonstrated in issues such as vote buying (Corstange, 2010; Gonzalez-Ocantos et al., 2012), racism (Kuklinski et al., 1997), turnout (Holbrook & Krosnick, 2010), and support for female candidates (Matthew J Streb, 2008). Research on these issues has shown that when questioned directly, respondents tend to underreport socially undesirable attitudes.
This could also be the case in the electoral punishment of corrupt candidates. Confronted with a direct question, respondents usually report a very critical view of corruption (Afrobarometer Round 36; World Values Survey Wave, 6, V2027). At the same time, people report very low levels of vote intention for corrupt candidates (Muñoz et al., 2016; Weitz-Shapiro & Winters, 2016; Winters & Weitz-Shapiro, 2013). Many empirical analyses with real election data, however, report that actual electoral punishment of corruption cases is limited (Chang et al., 2010; Costas-Pérez et al., 2012; Dimock & Jacobson, 1995; Eggers & Fisher, 2011; Golden, 2010; Peters & Welch, 1980; Reed, 1996; Rivero & Fernández- Vázquez, 2011).
This paradox could be explained by social desirability bias: interviewees may prefer to provide a more socially acceptable answer rather than their true opinion. This hypothesis suggests that people may not openly admit that they would vote for a corrupt candidate in a survey but they may do so when the behavior is not visible, in the real election, because privately, corruption is not such a big concern.
6 As reported in Chapter 1, a clear majority of respondents across all countries said that it is wrong and punishable if a government official (i) “gives a job to someone from his family who does not have adequate qualifications,” (ii) “demands a favor or an additional payment for some service that is part of his job,” or (iii) “decides to locate a development project in an area where his friends and supporters lived” (Q58a-c). 7 A clear majority of respondents across all counties consider it is never justifiable “someone accepting a bribe in the course of their duties” (V202).
54 Chapter 2
This is a common interpretation in recent literature on corruption voting. For instance, when arguing for the need of direct measures of electoral outcomes instead of survey statements, Chong et al. (2013) remark that “self-reported voting behavior suffers from social desirability bias.” Other authors have also expressed similar concerns Konstantinidis & Xezonakis (2013); Winters and Weitz-Shapiro (2013); Domínguez and McCann (1998). In the same vein, when making sense of the differential results regarding tolerance towards corruption in field experiment versus survey experiments, Boas, Hidalgo, and Melo (2018) argue that these are due to the strong social norms against corruption.
2.3 The tradeoff argument
An alternative explanation for this paradox is that survey questions about intentions of voting for corrupt candidates lack external validity. Voters might truly abhor corruption and consider it a liability when they evaluate candidates. But this is compatible with lower actual punishment in an election if other considerations influence the voting choice decision (Muñoz et al., 2016; Rundquist et al., 1977), or if voters do not receive or trust the information received on corruption (Ferraz and Finan 2007; Weitz-Shapiro and Winters 2015).
As shown in studies based on observational data, integrity is indeed a valued characteristic in a politician (Allen, Birch, and Sarmiento-Mirwaldt 2016; Birch and Allen 2015). However, as integrity only commands relative importance, voters have to consider many other aspects when it comes to casting votes in a real election. Therefore, the gap observed between what respondents report in surveys and what their eventual actions are in real elections could be attributed to the inability of surveys to gauge the tradeoffs that citizens face in real elections. According to this argument, direct questions in surveys do not fail due to their inability to uncover truthful answers, but due to their problems in estimating the relative importance of certain issues.
The literature has highlighted the importance of partisanship (Eva Anduiza, Gallego, and Muñoz 2013; Solaz, De Vries, and de Geus 2018), issue preferences
55 Social desirability bias
(Rundquist, Strom, and Peters 1977), or the perceived competence or economic performance of politicians (Esaiasson et al., 2014; Muñoz et al., 2016) in moderating the punishment of corruption in elections. I argue that in order to elicit correct estimates, questions in surveys have to be formulated that consider at least some of these other valued aspects.
2.4 Empirical strategy: social desirability bias
To assess the degree of social desirability bias in vote intention for corrupt politicians, this chapter uses a list experiment embedded in an online survey that took place in Spain in January, 2019. The data were collected within the project “LIMCOR: Limits to political corruption” (Fundació La Caixa 2016 ACUPO177). The Internet-based survey was completed by 1200 Spanish citizens between 18 and 89 years of age. The sample included quotas for gender, education and age in order to resemble the Spanish population. Refer to Chapter 1, section 1.5.1. for a justification of the use of data from Spain.
The disagreement between respondents’ assessments of corruption and their vote intention on the one hand, and the real electoral consequences of corruption on the other hand could be due to social desirability affecting survey instruments. If we could offer the respondents an instrument where they do not feel compelled to provide a social desirable answer (“punish corrupt candidates”), then we would likely find more people accepting the possibility of voting for a corrupt candidate.
List experiments provide a way to reduce the incentives of the interviewee to hide socially undesirable answers by asking the question in an unobtrusive way and explicitly assuring anonymity (Corstange, 2009). List experiments have been successfully used to remove measurement biases from sensitive issues such as racism (Kuklinski et al., 1997), self-reported voter turnout on telephone surveys (Holbrook & Krosnick, 2010), vote buying (Corstange, 2010; Gonzalez-Ocantos et al., 2012), and support for a female presidential candidate (Matthew J Streb, 2008), among others. If the mechanism behind the discrepancy between real
56 Chapter 2
data and survey responses in corruption voting is due to desirability bias, a list experiment should reveal it8.
Respondents were randomly assigned to three different groups. One group was assigned to a direct question and two groups were assigned to the list experiment, either the control or treatment group. In both experimental groups, respondents were presented with a list of plausible reasons for not voting for a mayoral candidate, and then were asked how many of these were good arguments for not supporting a candidate of their preferred political party, without having to specify which ones. The list was the same for both groups except for the sensitive item (being under investigation for corruption) that was only added to the list of items for the treatment group. This technique allowed respondents to freely answer without fearing that the interviewer or the researcher might disapprove of the answers, as one cannot know what items the respondent is referencing (Kuklinski et al., 1997). A posteriori, the difference in the means between the control and treatment group allow the estimation of the prevalence of the sensitive item in the treatment group.
Given that condoning corruption is only realistic in cases where the candidate possesses other valued characteristics, the candidate’s partisan label was matched to the respondent’s preferred party. Partisanship has been found to decrease the severity of the assessment of corruption cases (Anduiza, Gallego, and Muñoz 2013; Dimock and Jacobson 1995) and to moderate the electoral punishment of wrongdoings (Breitenstein, 2019; De Vries & Solaz, 2017). Furthermore, the candidate was described as a well preforming politician positively valued by the municipality’s residents.
8 A recent study draws attention to the possibility of failure of list experiments due to their length and complexity (Kramon & Weghorst, 2019). I reduce the possibility of breakdown by running the list experiment embedded in an online survey in a highly educated country. As internet-based surveys are self-administered, respondents can take the time they need to understand the question, consequently reducing possibilities of misunderstanding the instructions. Furthermore, I reduce complexity by keeping the list between 4 (control list) and 5 items.
57 Social desirability bias
The experimental question was as follows:
Imagine a candidate for mayor of your municipality who has extensive management experience, that residents value positively and who belongs to [respondent’s preferred party]. HOW MANY of the following items do you consider to be valid reasons not to vote for this candidate of [respondent’s preferred party]
The candidate…
The list of items for the control group (without the sensitive item) includes:
• was convicted in the past for sexually assaulting his secretary • is very religious against abortion and gay marriage • slightly stutters • is an atheist and supporter of removing religion from primary education
The list of items for the treatment group was exactly the same list but including:
• is under investigation for the urban redevelopment of land owned by a relative
Items were chosen to comprise a large range of circumstances, so that these could be either accepted or rejected according to different arguments. Furthermore, following the logic put forward by Glynn (2013), the list includes a negative correlation between two items so that respondents will only choose one of both. Respondents that like a religious politician that is against abortion and gay marriages, are also more likely to be in favor of obligatory religious education in school, and vice versa. This facilitates that each respondents at least selects one item and leaves without selecting another item. By limiting the possibility of ceiling and floor effects I ensure the unobtrusiveness of the list experiment (Janus, 2010). According to the results, the potential of ceiling and floor effects was minimal. In the control list, only 5% answered “none” and 8% answered “all of them”.
The wording of the corruption item was designed to closely match the actual cases. Most corruption scandals in recent years in Spain have taken place at the municipal level, and often related to the housing construction and urban planning. The item states that the candidate is under investigation, and not
58 Chapter 2
already convicted because this is the most common situation they face at the polls.
In order to match respondents’ and candidate’s partisanship, party identification was assessed in a previous question, “For which of the following parties do you feel more sympathy, or which one do you feel is closer to your own ideas?”. I they answered they did not feel close to any party they were asked a follow-up question asking what party they would select if they had to choose one.9 The dependent variable was integrated in the treatment condition, as it was the number of features in which the respondent wouldn’t vote for the candidate. I performed a randomization check (reported in Appendix A, Table A.1) that does not show any special problem in how the random assignment of respondents to treatment groups was performed.
The direct question was formulated in the same way as the list experiment, but instead of asking in how many cases one would not vote for the candidate, respondents were asked directly what they would do in the next elections. They were given three options: to vote for the corrupt candidate, to vote for another candidate of a different party, or to abstain.
The direct question was as worded follows:
Imagine a candidate for mayor of your municipality who has extensive management experience, that residents value positively and who belongs to [respondent’s preferred party]. The candidate is under investigation for the urban redevelopment of land owned by a relative. What would you do in the next election?
• I would vote for him/her • I would vote for a candidate of another party • I would not vote
9 21,4% of the respondents declared that they did not feel close to any of the Spanish parties. In this case instead of showing respondents the label of a party they would see the words “to your favorite party”.
59 Social desirability bias
2.5 Results: social desirability bias
The experiment does not reveal any hidden corruption voting due to potential social desirability. Using an unobtrusive way of inquiring does not elicit support for candidates under corruption allegations. As shown in Table 2.110, introducing the treatment item to the list results in a strong and significant increase of the dependent variable (0.825). This means that 82.5% of the respondents wouldn’t vote for a candidate under investigation for corruption, while 17.5% would do so. When I only take into consideration those that responded feeling close to a party and were therefore shown a candidate of their preferred party, the support for the corrupt politician increases substantially, as up to 26% of these respondents would vote for the candidate investigated for corruption.
These results conform to the results I obtain when the question was asked directly, as up to 23% of respondents accept that they would vote for the corrupt candidate even when asked undeviatingly. When only taking into consideration partisans, up to 28% would vote for the corrupt candidate of their preferred party. As a matter of fact, the support for corrupt politicians is even higher when respondents were directly asked what they would do. In Table A.3 in the Appendix A, I provide evidence that the experiment does not suffer from design effects that could affect the validity of the experiment and the results (see Blair and Imai 2012 for more information regarding this test).
Table 2.1. Results across different treatment conditions
Control Treatment Difference % Vote for % Voter for list List corrupt - list corrupt – direct
All 2.02 2.845 respondents (0.046) (0.060) 0.825 17.5% 22.75%
Only 1.96 2.70 partisans* (0.052) (0.069) 0.74 26% 28.27% *Only respondents that answered feeling close to a party and were therefore assigned their favorite party in the list experiment (i.e. excluding those that answered not feeling close to any party).
10 See table A2 in Appendix A for the piecewise proportions.
60 Chapter 2
The results do not provide support for the idea that direct survey questions on voting for corrupt politicians suffer from social desirability bias. If that were the case, we would have observed a higher acceptance or ‘forgiveness’ of corruption in responses to an unobtrusive measure like the one included in the list experiment. On the contrary, results show that a very similar proportion of respondents accept voting for a corrupt politician both in the unobtrusive question as in the direct question.
These outcomes suggest the need to identify alternative explanations when trying to make sense of the discrepancies between survey responses and actual behavior with regard to corruption voting. Results from the list experiment do not support the idea that voters have a true structure of preferences in which corruption is unimportant in their vote choice that gets masked by social desirability concerns in survey research. Voters seem to genuinely dislike corruption and give weight to it in their choice function.
The fact that election results tend to show limited punishment for corruption allegations (and thus contradict survey responses that show great willingness to punish corruption at the polls) should not be attributed to social desirability issues without further research. Such a result is compatible with alternative explanations of corruption voting. Voters might dislike corruption, but this is not the only or most important consideration they take into account when deciding their vote. Partisanship, policy proximity, or competence might be traded off against integrity and lead to vote for candidates despite them having been accused of corruption. In the next section I assess whether the discrepancy between what is observed in surveys and that observed in real elections is due to tradeoff voters’ face in real elections.
61 Social desirability bias
2.6 Empirical strategy: tradeoff argument
This section assesses whether the differences between what we usually observe in surveys and what we actually observe in real elections is due to the tradeoffs that citizens face when casting a vote. According to this argument, standard survey questions are not doing a good job at estimating the relative importance of corruption.
In order to test this argument, I randomly assigned respondents two different questions: one simple question and another where the tradeoff is emphasized. In the simple question, respondents were asked what they would do if the candidate running for their favorite party was under investigation for corruption. The response options were the same as in study 1 (I would vote for him/her, I would vote for a candidate of another party, I would not vote). Please note that in this question I actually already include a tradeoff, as respondents are faced with a corrupt candidate from their favorite party. This is the case because the query addressed in this chapter − whether voters would punish a politician for corruption − only makes sense if voters already have a certain predisposition to vote for this politician. If from the very beginning voters have no intention of voting for a certain politician, they have no room to punish the politician if they find out about her wrongdoings.
The second question was formulated by emphasizing the tradeoff more11. In this question, the candidate was not only a member of the respondent’s favorite party, but was also described as having extensive management experience and as being positively valued by citizens. The tradeoff I included in this question was inspired by previous research that found that partisanship and a politician’s competence mitigate the punishment of a corrupt politician at the elections (Eva Anduiza, Gallego, and Muñoz 2013; Muñoz, Anduiza, and Gallego 2016; Solaz, De Vries, and de Geus 2018; Breitenstein 2019). The justification for using this type of corruption is the same as for Study 1.
11 This question is the same as the direct question used in Study 1.
62 Chapter 2
Wording of the simple question:
Imagine a candidate for the mayor of your municipality that belongs to [respondent’s preferred party]. The candidate is under investigation for the urban redevelopment of land owned by a relative. What would you do in the next municipal elections?
Wording of the tradeoff question:
Imagine a candidate for mayor of your municipality who has extensive management experience that residents value positively and who belongs to [respondent’s preferred party]. The candidate is under investigation for the urban redevelopment of land owned by a relative. What would you do in the next elections?
This study was also implemented in Spain, following the same reasoning put forward in Chapter 1, and to facilitate comparing the results of both studies. In this case, the internet-based survey was implemented in May 2019 and was completed by 1748 Spanish citizens who were between 18 and 65 years of age. The sample included quotas for age, gender, education and region of residence to reflect the Spanish population.
2.7 Results: tradeoff argument
Table 2.2 shows the percentage of respondents that express that they would vote for the corrupt candidate across the different questions. More respondents express a willingness to overlook corruption when the question is formulated as a tradeoff. In this case, up to 20% would vote for the corrupt politician (24.5% if I only consider partisans), whereas when posed the simple question, only 14% (or 18% of the partisan respondents) express willingness to vote for the corrupt politician.
In order to test whether the observed differences across the simple and the tradeoff question are statistically significant, I regress vote intention on a dummy variable for the group that answered the tradeoff question, using a multinomial
63 Social desirability bias
logistic model. The reference outcome is voting for an alternative candidate. Table 2.3 shows that respondents that were presented with the tradeoff have a significantly higher probability of expressing that they would vote for a corrupt candidate compared to those that received the simple question. The result holds when a set of control variables is included in the model (reported in Table A.4 in Appendix A). These results support the tradeoff argument. The extremely high willingness to punish corruption at the polls that citizens express in surveys is substantially moderated when they are reminded of the tradeoffs they may face in real elections.
Table 2.2. Percentage of respondents that express that they would vote for the corrupt candidates across the different type of questions
Simple question Tradeoff question
All respondents 13.89 20.24 Only partisans 17.91 24.5
Table 2.3. Multinomial logistic regression for tradeoff argument
Vote for corrupt Abstain Reference: Simple question Tradeoff question 0.56*** (0.17) 0.24 (0.13) Constant -1.25*** (0.13) -0.25** (0.09) chi2 11.66 p 0.00 Observations 1166
Note: Standard errors in parentheses. Dependent variable: vote for corrupt, vote for alternative or abstain. Base category: vote for alternative candidate * p<0.05, ** p<0.01, *** p<0.001
64 Chapter 2
2.8 Conclusions
According to the list experiment results, social desirability bias does not explain the discrepancy between the reported rejection of corruption in surveys and the actual punishment of corruption in elections. When people express a negative evaluation of corruption cases, they seem to report these attitudes sincerely. Nevertheless, when it comes to real elections, voters have to consider additional aspects beyond a politician’s integrity, and they may trade integrity for other characteristics that they value. The discrepancy between the low vote intentions for corrupt politicians in surveys and the limited punishment of corruption commonly observed in real elections is probably a consequence of the inability of standard survey question to estimate the relative importance of corruption in elections. The results show that the probability of expressing support for a corrupt candidate increases significantly when the survey question aims to reflect these tradeoffs.
These findings have implications both for our substantive understanding of corruption and its electoral consequences, and for the empirical analysis of how corruption affects voting choice. In substantive terms, these findings contribute to the understanding of the limited electoral consequences of corruption. Corruption is only one of the many factors that people weigh up when deciding who to vote for. Voters may truly abhor corruption, but even if they dislike corrupt politicians, under certain circumstances they are prepared to consider voting for them, because they may value other characteristics that they have.
Empirical researchers should take this into account when conceiving survey instruments to estimate the electoral effects of corruption, and when interpreting their results. Survey instruments should present situations where the choices mirror this complexity. My findings suggest that excessively simplistic scenarios, where respondents are not presented with the tradeoffs that exist in reality, will be unlikely to provide valid information about how people punish electoral corruption at the polls (or not). If this is not ensured, one should not be surprised that the estimates of the ceteris paribus punishment of corruption do not mirror the levels of punishment observed in real elections, where voters do not face a ceteris paribus situation but a choice among a limited set of options that differ from each other in many, potentially relevant dimensions.
65
Chapter 3
Tradeoffs and corruption
3.1 Introduction 12
The previous empirical chapter provided evidence that social desirability does not explain the discrepancy between the reported rejection of corruption in surveys and the actual punishment of corruption in elections. According to my findings, when people express a negative evaluation of corrupt politicians, they are reporting these attitudes sincerely. My argument is that in real elections, voters consider additional aspects beyond the politician’s integrity, and they may trade off integrity for these other qualities. According to this argument, direct questions in surveys do not fail due to their failure to uncover truthful answers, but due to their problems in replicating the complexity of decision making in real elections.
This chapter aims to further assess this argument and the tradeoff hypothesis. The tradeoff hypothesis proposes that voters condone corruption when politicians possess other valued characteristics. While some studies have looked
12 A version of this chapter is published as: Breitenstein, Sofia. 2019. “Choosing the Crook: A Conjoint Experiment on Voting for Corrupt Politicians.” Research & Politics https://doi.org/10.1177/2053168019832230
66 Chapter 3
at the moderating effects of (co-)partisanship (e.g. Eva Anduiza, Gallego, and Muñoz 2013; Solaz, De Vries, and de Geus 2018), others have assessed the conditioning effects of the candidate’s economic performance (e.g. Esaiasson et al., 2014; Zechmeister & Zizumbo-Colunga, 2013). By providing respondents with different pieces of information, this chapter creates a multidimensional decision making process and in this way manages to test both tradeoff hypotheses in one experimental setting. It consequently increases the external validity of the study, as it provides a more accurate account of what is happening in real elections, where voters are confronted with multiple tradeoffs when casting their votes. In addition, the chapter discusses two ways of understanding the tradeoff hypothesis and the differing impact these could have on anti- corruption policies.
By exploiting an original conjoint experiment, this chapter provides compelling evidence of the relative importance of corruption when casting a vote and the mitigating effects of other valued characteristics in candidates. Even when obtaining highly credible information, partisanship determines the vote to the same extent as corruption. Additionally, co-partisanship and strong economic performance moderate the negative effect corruption has on voting choice.
3.2 Theoretical framework
Regarding the reasons that informed and free citizens vote for corrupt politicians, the literature has highlighted the relative importance of corruption when selecting a candidate. According to Rundquist et al. (1977), voters seem to care about corruption, but they also have other concerns and may trade integrity for other valued characteristics of the candidate. These authors pointed to an implicit exchange between the integrity of the candidate and her position on certain policies or issues. Voters might forgive corruption when malfeasant candidates take a position on issues that are more important to them.
A number of previous studies have shown that partisanship moderates the perception of corruption (e.g. Anduiza et al., 2013; De Vries and Solaz, 2017; Eggers, 2014). In a survey experiment conducted in Spain, Anduiza et al. (2013)
67 Tradeoffs and corruption
found that respondents consider corruption less severe when it affects the party they feel closer to. Nevertheless, Konstantinidis and Xezonakis (2013) did not corroborate these results in a study conducted in Greece. The authors of the latter study argued that the different results might be due to the strong partisan alliances in place in the political context where the survey experiment by Anduiza et al. (2013) was launched. In this chapter, I reassess these results in the same country, but during a time when new parties have recently entered the political arena and, therefore, party loyalties might not yet be as strong.
Moreover, this chapter poses a harder test for the partisan tradeoff. Respondents also received varying information regarding other candidates’ attributes that have been found to be important determinants of the vote, such as the economic performance of the candidate (for a review, see Lewis-Beck and Stegmaier, 2007). Building on the same idea of the relative importance of corruption, recent studies set their attention on a tradeoff between integrity and the competence of candidates in other areas. According to this literature, voters disregard corruption when candidates deliver other benefits, such as economic growth or other public goods. Several prior studies on voting for corrupt politicians provide evidence that corruption is less punished in good economic contexts (Klašnja and Tucker, 2013; Zechmeister and Zizumbo-Colunga, 2013), when the politician has a good management record (Esaiasson et al., 2014; Muñoz et al., 2016), or when he has implemented favorable economic policies (Konstantinidis and Xezonakis, 2013). Nevertheless, other research did not obtain the same results. According to Winters and WeitzShapiro (2013), when voters learn about corruption, they punish those candidates even if they performed well. Therefore, it is essential to reassess this hypothesis and verify if the economic performance of a candidate actually determines whether voters condone corruption at the polls. Likewise, this study poses a hard test for this hypothesis as respondents also received information on the candidates’ partisan affiliations.
This chapter provides at least two contributions to the literature. It is one of the first studies to test both tradeoffs in one survey experimental setting. By using the data of an original conjoint analysis, this study evaluates how respondents make decisions in a multidimensional scenario where they have to take several tradeoffs into consideration. Whereas most previous studies have tested the moderating effect of partisanship and a good economic performance separately,
68 Chapter 3
this chapter reassesses their moderating effect in a multidimensional scenario. Respondents received information on various candidates’ characteristics, thus posing a more difficult test for both hypotheses.
Currently, other researchers are assessing corruption accountability with a multidimensional perspective. However, the main aim of these studies differs substantially from this chapter, both in the hypotheses they are trying to assess and in their research design. While in this chapter I assess the moderating effect of candidates’ characteristics on the electoral punishment of corruption, Visconti and Mares (2018) and Klašnja et al. (2017) implemented conjoint experiments to mainly assess whether certain corruption characteristics moderate the negative effect of malfeasant activities. Perhaps the study by Franchino and Zucchini (2014) is most closely related to this chapter; however, this also differs considerably from this chapter in terms of the literature it engages with and the candidate characteristics that it assesses (refer to Appendix B, section B.6 for a detailed comparison between this chapter and the other papers).
In addition to assessing the conditional effect of co-partisanship and the economic performance of a candidate, this study also estimates and compares their relative weight on the likelihood of voting for a politician. Participants were presented with profiles of two mayors with randomly assigned information on the candidates’ party affiliation, economic performance, integrity (corruption), educational and managerial qualities, and gender, and were asked to report their probability of voting for each candidate. This allows for determining the relative causal effect of each candidate’s characteristics on the respondent’s probability of supporting the candidate (Hainmueller et al., 2014).
The distinction between the relative importance of different factors and their moderating effect on the electoral punishment of corrupt candidates is directly linked with the second contribution of this chapter. Previous research has not agreed on exactly what the tradeoff hypothesis entails, and is, therefore, unclear on how to measure it. Some studies have looked at the factors that moderate the electoral punishment of corruption, while others have theorized about the relative importance of corruption. Nevertheless, these two ways of explaining the tradeoff hypothesis entail a different causal mechanism and, therefore, may
69 Tradeoffs and corruption
entail a different consequence for policies that aim to combat corruption by increasing citizens’ awareness.
The relative weight argument, drawn from the discussions in previous literature and the public arena, proposes that corruption is just another factor voters consider when casting their vote and that they might give more importance to issues other than corruption (e.g. Fisman and Golden, 2017). This explanation is in keeping with the argument of Rundquist et al. (1977), which assumes a rational voter that weighs the candidates’ characteristics and votes accordingly. According to this hypothesis, voters choose a crooked politician when they weigh her partisan affiliation or her economic performance stronger than her integrity. The second approach, which I call the “conditional punishment argument”, proposes that the negative effect of corruption on the support of a candidate may diminish when the candidate exhibits other positive characteristics. This argument is in keeping with the research, showing that a good economic situation moderates the negative evaluation of a corrupt candidate (e.g. Zechmeister and Zizumbo-Colunga, 2013). In this case, the causal mechanism could be less rational, as voters might vary how they weigh a candidate’s corruption allegations depending on other candidate qualities. Therefore, the assumption here is not that voters rationally choose a corrupt candidate because they give priority to other characteristics. In this case, voters could be equally or more concerned with the candidate’s integrity; nevertheless, other positive candidate characteristics could unconsciously influence integrity’s relevance on their decision-making. For example, Anduiza et al. (2013) provide evidence that respondents diminish the perceived severity of a co-partisan’s corrupt activity to avoid cognitive dissonance.
Both hypotheses have different substantive implications. While the relative weight hypothesis conceives a rational voter that simply chooses to overlook corruption, the conditional punishment hypothesis conceives a voter that might be unconsciously driven by psychological biases. Moreover, these two conceptions have different implications for anti-corruption campaigns. If voters are rational and vote according to the importance they attach to each candidate’s characteristics, anti-corruption campaigns simply have to inform about politicians’ malfeasant behavior, insist on the negative consequences of corruption for society, and address the necessity for voters to hold politicians accountable. However, if voters are driven by unconscious biases, the strategies
70 Chapter 3
that motivate them to hold politicians accountable at the polls might not be as straightforward and successful. In that case, even if voters are informed about the negative consequences of malfeasant behavior and their ability to hold politicians accountable, psychological biases, driven by co-partisanship or a strong economic performance, might affect how they perceive the severity of malfeasant activities of their preferred politicians.
3.3 Empirical strategy
To test the tradeoff hypotheses, this chapter uses the data of an original conjoint experiment embedded in a representative survey of the Spanish population. Survey experiments have proven to be a unique technique to assess causal inference and to study individual political behavior (see Chapter 1, section 1.5.2 for a review on advantages and limitations of survey experiments). In conjoint experiments, in contrast to standard survey designs, several pieces of information are manipulated in one setting. It consequently allows for varying and analyzing different dimensions of the studied phenomenon and so increases the external validity of the research.
The experiment was embedded in an online survey (N=2275) conducted in Spain in June 2016 with Qualtrics software. The sample was comprised of Spanish citizens aged 18 and older and included sex, age and education quotas in order to achieve an accurate representation of the Spanish population. Table 3.1 shows a comparison of the socio-economic demographics of my survey with a face-to-face survey carried out in the same month by the official Spanish Statistical Office (CIS) on a representative sample of the Spanish population. The sample of the experiment is slightly younger and more educated than the CIS sample. Nevertheless, as the treatment was assigned through a complete randomization and the aim of this chapter is to measure causal relationships, a fully representative sample is not required.
71 Tradeoffs and corruption
Table 3.1. Characteristics of the samples
Experiment CIS3142 Age (average) 40.9 49.8 Gender (% women) 48.3 51.4 Education Primary education or less 5.3 24.3 Secondary education and upper secondary 58.3 53.9 Tertiary education 36.4 21.3 Ideology (average 0-10) 4.3 4.1 Close to PP (%) 14.2 Close to PSOE (%) 15.3 Close to Podemos 19.4 Close to C's (%) 12.8 Satisfacion with democracy (average 1-10) 4.1 Political sophistication 1.9 N 2,275 2,484
In order to check whether the randomization of the treatments was successful, a multinomial logistic regression comparing individual socio-economic demographics across the different groups is provided in Table 3.2. As the chi square is not significant, we can be sure that the randomization was successful. The randomization test was made for both the samples with and without runners. In the chapter I only present the results of the sample without runners, the tenth percentile that took the least amount of time to answer each round was dropped. This corresponds to less than 15 seconds to answer in the first round, 10 seconds in the second and 9 in the third. Less time than this would certainly not be enough to read the whole experiment and answer the questions. In total, 1198 observations were deleted; the data shown in the chapter are drawn from 12,288 evaluated profiles or 6,144 pairings. As discussed in the next section, respondents participated in three rounds and in each round were presented two profiles. The results with runners do not change substantially, and they can be made available upon request.
72 Chapter 3
Table 3.2. Randomization Test. Mlogit Regression Model. Dependent Variable: Corruption Treatment
Without runners With runners Accused Accused Accused Accused parties judge parties judge Gender -0.02 (0.05) 0.07 (0.05) -0.02 (0.04) 0.08 (0.04) Age -0.00 (0.00) -0.00* (0.00) -0.00 (0.00) -0.00 (0.00) Education 0.02 (0.01) 0.03* (0.01) 0.02 (0.01) 0.03* (0.01) Income 0.02 (0.01) 0.02 (0.01) 0.02 (0.01) 0.01 (0.01) Unemployed 0.07 (0.06) 0.04 (0.06) 0.09 (0.06) 0.04 (0.06) Ideology 0.00 (0.01) -0.00 (0.01) -0.00 (0.01) 0.00 (0.01) Political sophistication -0.00 (0.02) 0.01 (0.02) -0.01 (0.02) 0.00 (0.02) (Unsatisfied with democracy) Neither satisfied nor unsatisfied with democracy -0.03 (0.07) -0.11 (0.07) -0.06 (0.06) -0.12 (0.06) Satisfied with democracy -0.11* (0.05) -0.12* (0.05) -0.10 (0.05) -0.13* (0.05) (No party id) Party id: PP 0.15 (0.08) 0.12 (0.08) 0.13 (0.08) 0.08 (0.08) Party id: PSOE 0.06 (0.08) -0.03 (0.08) 0.02 (0.08) -0.07 (0.08) Party id: Podemos 0.04 (0.08) -0.00 (0.08) 0.04 (0.07) -0.00 (0.07) Party id: Ciudadanos 0.07 (0.08) -0.05 (0.08) 0.05 (0.08) -0.05 (0.08) Other party id 0.03 (0.08) -0.03 (0.08) -0.01 (0.07) -0.05 (0.07) Constant -0.14 (0.13) -0.10 (0.13) Chi2 36.71 35.12 P 0.13 0.17 Observations 12010.00 13188.00
73 Tradeoffs and corruption
3.3.1 Design and wording of the experiment
Respondents were presented the profiles of two mayors with a set of five attributes with independently randomly assigned categories (or components) and were asked to report their likelihood of voting for each candidate if they were running for elections in their hometown. Each respondent was asked to repeat the same task three times with random and varying pairs of candidates. The experiment was completely randomized, so no combination of attributes was restricted; the sequences of the attributes were also randomized across each respondent but kept constant over the three tasks. This procedure allows for the estimation of the relative influence of each attribute and enables us to assess how the attributes interact with each other (Hainmueller et al., 2014). Hence, the conjoint design is ideal to evaluate what is the variable that most determines the vote (relative weight argument) and whether any factors moderate the negative effects of corruption (conditional punishment argument).
Table 3.3 shows the categories for each attribute and the corresponding text that was shown in the experiment.
Table 3.3. Attributes and text for each component
Female Sex Male PP PSOE Party Podemos Ciudadanos Has compulsory education and little management experience Qualities Has university education and prolific management experience Economic Investments in the municipality have increased so that unemployment has decreased by 5 % performance Investments in the municipality have decreased so that unemployment has increased by 5 % Has been characterized for his/her honesty Corruption Has been accused by other parties of corruption for awarding contracts in exchange for gifts Has been accused by the judge of corruption for awarding contracts in exchange for gifts
74 Chapter 3
The design presented in this chapter aims to increase the external validity of the experiment while avoiding the practice of confronting respondents with a too demanding task. I tried to find the right balance between reflecting the multidimensional scenario of real elections and simplifying the information in order to not over-challenge respondents. For each mayor, respondents were presented with five attributes or characteristics with different, randomly assigned categories per attribute. To achieve full randomization, the attributes included in the experiment were the following: information about corruption (labelled as mandate in the vignette presented to respondents to avoid social desirability), party affiliation, economic performance (labelled as outcomes to avoid social desirability once again), educational and managerial qualities and gender.
The information about corruption was randomized over three categories: honest, accused of corruption by other parties and accused of corruption by a judge. In order to operationalize the information about corruption, I have used a similar approach as Weitz-Shapiro & Winters (2015). The credibility of the information has been varied through the credibility of the source issuing the information. Other parties have been employed as a source with low credibility, while the judge has been used as a source with high credibility. In the pilot version of the experiment, I also included a category of corruption without a reference to the source of the information to have a condition where the uncertainty of the information is completely minimized, so as to evaluate whether some respondents do not trust the information provided by the judge. According to the results of the pilot, respondents trust the information provided by the judge and the one without a source equally. For that reason, I have decided not to include this category in the final version of the experiment.
As far as the categories of party attachment are concerned, the four most voted parties in the election of December 2015 were selected: PP (conservatives, currently in government), PSOE (social-democrats), Podemos (left) and Ciudadanos (center-right liberals). These parties, apart from being the parties with most support, are also spread out on the ideological spectrum and therefore represent a heterogeneous set of preferences. For the analysis, I used variable partisanship as measured by asking respondents what political party they feel close to before the experiment and then assess whether it corresponds with the party of the candidate that is being evaluated. This variable has three categories: being co-partisans (same party), different party (respondent is partisan of
75 Tradeoffs and corruption
another party than the one seen in the experiment) or no party identification (respondent declares not feeling close to any party but assesses candidates with party attachments).
Economic performance was also randomized across two values: strong economic performance and weak economic performance. Economic performance was operationalized by following two strategies. First, I mentioned in general terms that the municipality has attracted investments (or not), this prevents linking the economic performance to the candidate as there is not a clear expression that he/she attracted investment. I opted for this wording because I wanted to see whether the tradeoff of integrity against economic performance is due to the importance respondents assign to obtaining good economic outcomes regardless of the other qualities of the candidate (operationalized in another treatment). If respondents only care about getting good economic outcomes, no matter if it is due contextual matters or due to corrupt activities, then the positive effect of good outcomes should hold even when the candidate is corrupt and has low qualities.
Second, I have made references to the unemployment rate, which together with corruption and the economic situation are the three most important problems of Spain mentioned by respondents in the CIS barometer. Referring to unemployment ensures that the respondents are familiar with the issue and care about it. Furthermore, the analysis of electoral consequences in 32 European countries done by Bågenholm (2013) has shown that unemployment rates are more strongly correlated to incumbents’ electoral performance and general government change than allegations and corruption scandals. As unemployment is a very salient issue in Spain and has been proven to be an important factor in the electoral consequences of incumbents, using this factor ensures a strong treatment that poses a hard test for the partisan tradeoff hypothesis.
Educational and managerial qualities of the politicians took two values: low qualities and high qualities. In order to operationalize a candidate with high and low educational and managerial qualities I have used a combination of traits that are selected as the most important qualities a politician must have in the surveys run by the CIS (CIS: 2905 and CIS: 2930). In both studies the most valued attribute of a politician is honesty/integrity by a large margin. As this quality is already treated in the attribute with information about corruption (‘Mandate’), I
76 Chapter 3
have included the other traits respondents consider important. These are a combination of education and technical knowledge, management skills, efficiency, etc. Furthermore, education is regularly used by economists as a quality signal. Finally, gender was randomized across the categories male or female.
All the information provided to the participants was approved by the Ethical Committee of the Autonomous University of Barcelona (ref: CEEAH: 3323). Before starting the experiment, it was clearly stated that the information referred to a hypothetical scenario and that participants could abandon the survey at any time. After the experiment participants obtained debriefing information and contact details.
Concerning the dependent variable, respondents expressed their probability of voting for each candidate on a scale from zero (“would never vote for”) to 10 (“would definitely vote for”). The answers were rescaled from zero to one. I selected this dependent variable instead of a forced choice between candidates because the probability of voting allows respondents to not vote for any candidate and, therefore, bears a stronger resemblance to real elections (Refer to Appendix B, section B.5.3 for the results of the forced choice as a dependent variable). Furthermore, Hainmueller et al. (2015) show that paired conjoint experiments with a question for each option are better at eliciting behavior in real situations. Overall, there were 12,284 evaluated profiles or 6142 pairs of candidates. Table 3.4 shows the distribution of the vote probability across the different treatments. The average vote probability of an honest candidate is 0.49.
Table 3.4. Descriptive of Dependent Variable: vote probability
Treatment Mean Std Dev Obs Honest 0.49 0.35 4189 Accused by parties 0.27 0.3 4032 Accused by judge 0.22 0.29 4063 Total sample 0.33 0.34 12284
77 Tradeoffs and corruption
3.4 Results
The probability of respondent i voting for a candidate k in task j is modeled as a function of the candidate’s (co)partisanship13, economic performance, integrity, and other characteristics:
Pvijk= β1*Partisanshipijk + β2*Economic performanceijk + β4*Corruptionijk + β5*Otherijk + eijk
According to the relative weight model, voters elect corrupt candidates when they weigh other issues or characteristics more strongly than corruption. Consequently, this hypothesis is corroborated if the effect of other variables, such as partisanship and economic performance, is stronger than the effect of corruption on the probability to vote for a candidate.
The complete randomization of the experiments allows us to estimate the average marginal component effects (AMCEs) by fitting a linear regression and clustering for respondents, as each respondent repeated the same task three times (Hainmueller et al., 2014)14. The AMCEs (shown in Figure 3.1) can be interpreted as the marginal effects of changing a given characteristic (or category) on the population’s probability to vote for a candidate averaged over all possible values of the other characteristics. Results show that the information about corruption has a strong negative effect on support. The accusation of corruption by other political parties decreases the support of a candidate by 0.22 (on a scale from 0 to 1) as compared to the level of support for an honest candidate, while the accusation of a judge does so by 0.27. It is especially remarkable to observe that even though the experiment uses a strong treatment of corruption and refers explicitly to corruption, under certain conditions, partisanship has an equally strong effect on the support of a candidate as corruption15 (Refer to Appendix B for several robustness checks.) Seeing the
13 Measured by combining the respondent’s party identification, acquired before the experiment, and the party that is being assessed in the experiment. 14 Results are corroborated using a linear regression with individual fixed effects. 15 These results hold along all tests except for the model that uses the forced choice as a dependent variable. The relative effect of the variables is slightly different when respondents were forced to choose one of the two candidates.
78 Chapter 3
profile of a candidate belonging to a different party decreases the support to the same degree as an accusation of corruption by a judge. These results were corroborated using a variable that differentiates between nonpartisans and partisans of another party and another variable that measures party preferences instead of partisanship (see Appendix B, Figures B.2 and Table B.4).
Regarding the economic performance of the candidate, a weak economic performance has a significant negative effect on the support of a candidate, but this effect is considerably weaker than the effect of corruption. A low education and little management experience also significantly decrease the level of support for a candidate.
Overall, the results of the experiment show that respondents not only care about corruption, but there are other candidate characteristics that determine the likelihood of voting for that candidate. Indeed, a certain combination of variables increases the probability of voting for a corrupt candidate. For example, respondents’ average probability of voting for an honest candidate from a different party with a weak economic performance and low educational and managerial qualities is only 0.39, while the probability of voting for a co- partisan candidate with a strong economic performance and high educational and managerial qualities who is accused of corruption by other parties is 0.61.
79 Tradeoffs and corruption
Figure 3.1. Average marginal component effects (AMCE)
Woman
Different party
Low qualities
Weak performance
Accused parties
Accused judge
-.3 -.2 -.1 0 .1
To test the conditional punishment argument, it is necessary to determine whether the negative effect of corruption on the support of a candidate varies according to the partisanship or the economic performance of the candidate. To do that, I estimate the average difference in the AMCEs of corruption between different profiles of candidates (see Hainmueller et al., 2014: 12). The conditional punishment hypothesis is corroborated if corruption has a weaker negative effect when the candidate is a co-partisan or delivered a strong economic performance.
Table 3.5 shows the percentage of change in the probability of voting for a co- partisan candidate or a candidate with a strong economic performance when accused of corruption (grouping both partisan and judicial accusation of corruption). The results show that an accusation of corruption has a weaker negative impact on the likelihood of receiving a vote when the candidate belongs to the same party as the voter. Corruption decreases the probability of voting for a candidate belonging to a different party by 52%, while the vote probability only decreases by 40% for a co-partisan candidate. The semi-elasticities in Table 3.6, which provide the proportional change in Y for a change in X, confirm that this differential impact is strong and significant at a 99% confidence interval. As far as the economic performance of the candidate is concerned, the conditional punishment hypothesis only holds true to some extent. Information on
80 Chapter 3
corruption has a weaker effect when a candidate delivered a strong economic performance; nevertheless, the difference is small (2 percentage points) and only holds at a 90% confidence interval. In summary, the results of this experiment show that the negative effect of corruption is indeed attenuated by co- partisanship and, to a lesser extent, by the economic performance of the candidate.
Table 3.5. Predicted probabilities and the relative reduction of corruption
Same Party Different Party Confidence Reduction Vote Confidence Reduction Vote Interval of Vote Probability Interval of Vote Probability Probability Probability Lower Higher Lower Higher (%) (%) Honest 0.746 0.724 0.768 0.445 0.431 0.459 Corrupt 0.448 0.426 0.470 39.930 0.211 0.201 0.220 52.658
Strong Weak
Economic Performance Economic Performance Confidence Reduction Vote Confidence Reduction Vote Interval of Vote Probability Interval of vote Probability Probability Probability Lower Higher Lower Higher (%) (%) Honest 0.542 0.525 0.558 0.441 0.425 0.456 Corrupt 0.278 0.266 0.289 48.695 0.216 0.206 0.227 50.931
Note: The first column of each group shows the predicted values of respondents voting for a candidate. The second column shows the relative reduction in the probability of voting for the same candidate with the added component of being accused of corruption (grouping both partisan and judicial accusation of corruption), tacking always as a reference the probability of voting for an honest candidate in each group.
81 Tradeoffs and corruption
Table 3.6. Derivatives expressed as a semi-elasticity
ey/dx Contrast Partisanship Different party -0.764 (Reference) Same party -0.512 0.251*** Economic performance Weak performance -0.763 (Reference) Strong performance -0.699 0.064*
*** p<0.01, ** p<0.05, *p<0.1 Note: The first column shows the proportional change of Y when the candidate is corrupt across the different categories of partisanship and economic performance. The contrast columns show the difference in the change of Y between each reference category and the rest of the categories and the significance associated with these differences.
3.5 Conclusions
This chapter provides compelling evidence that respondents care about corruption but also care about other candidate characteristics. Analyzing the results of a conjoint experiment proves that in a multidimensional setting, participants trade out integrity for other valued characteristics. While previous studies have tested the moderating effect of partisanship and a good performance individually, this study contributes to the literature by assessing how respondents react when they also receive information on many other candidates’ characteristics. In addition, this study helps us to understand not only the moderating effect of partisanship and a strong economic performance, but also their relative importance on the probability of voting for a candidate.
The results of this multidimensional survey experiment confirm a tradeoff between integrity and co-partisanship. In line with previous findings (Anduiza et al., 2013; Barnes & Beaulieu, 2014; Beaulieu, 2014), co-partisanship strongly moderates the negative effect corruption has on the likelihood of voting for a politician. Furthermore, partisanship, together with corruption, is the attribute that most determines the vote. However, these results only partly corroborate a tradeoff between the economic performance and the integrity of a candidate.
82 Chapter 3
While in this experiment, corruption determines the vote to a much greater extent than the economic performance of a candidate, the economic performance does moderate the negative effect of corruption. However, this conditioning effect is mild and does not hold across all robustness checks. In summary, this study corroborates the importance of partisanship in condoning corruption. While this study finds some evidence in favor of the economic performance tradeoff, these results are by no means as strong as the partisanship tradeoff.
As shown in the analysis of this chapter, the relative weight and the conditional punishment model can be compatible, as partisanship determines the vote to the same extent as corruption; however, partisanship also moderates the effects of corruption. Although these models are not mutually exclusive, it is important to distinguish them in future research as they could entail different causal mechanisms based on distinct rationality in voters’ decision making. Ultimately, these models could have a differential impact on anti-corruption policies that aim to engage citizens in the control of corruption. Due to the properties of the conjoint design, in this chapter, I could not test the rationale behind respondents’ decisions. Future research should now determine the exact pathways that might drive each argument.
Concerning the elicitation of stated preferences with hypothetical scenarios, Hainmueller et al. (2015) demonstrated that paired conjoint experiments are highly successful in replicating the decision making that takes place in real settings. Furthermore, I consider the high credibility of the information provided in this experiment an advantage of this study as it poses a solid test for the tradeoff hypotheses. This chapter shows that even when obtaining highly credible information and, therefore, being certain that a candidate is corrupt, respondents are willing to trade integrity for other valued characteristics of the candidate. Hence, I provide clear-cut evidence that even informed citizens might choose to overlook corruption.
83
Chapter 4
Reassessing the competence corruption tradeoff
4.1 Introduction 16
Chapter 3 of this dissertation provides evidence that in a multidimensional setting, voters trade off integrity against other characteristics that they value in candidates. It also assesses some possible tradeoffs that citizens can face in elections. The results show that partisanship, together with corruption, is the attribute that most determines who a citizen votes for. Furthermore, co- partisanship strongly moderates the negative effect corruption has on the likelihood of voting for a politician. However, the results only partly corroborate a tradeoff between the economic performance and the integrity of a candidate.
This chapter reevaluates this hypothesis by assessing to what extent and why citizens might (or might not) favor politicians who are dishonest, but who deliver
16 This chapter is part of a work in progress: “Too Crooked to be Good: Tradeoffs and the Punishment of Malfeasance” (written together with Enrique Hernández)
84 Chapter 4
public goods and superior outcomes. Unlike Chapter 3 and previous studies, in this study I assess citizens’ evaluations and their likelihood of supporting politicians who manage to get the job done precisely because they bypass the established legal procedures or are openly corrupt. I therefore test whether citizens favor outcome accountability over procedural accountability in situations where there is a direct dilemma between the two options. That is, in circumstances in which politicians can either follow the established legal procedure and achieve a suboptimal outcome, or bypass that procedure in order to achieve an optimal outcome.
Assessing citizens’ reactions to situations that pose a dilemma between outcome and procedural accountability is not only relevant in and of itself to evaluate theories about corruption voting, but also because these tradeoffs might be present when evaluating politicians in elections. While political malfeasance has clear negative consequences in the long run, in certain situations, breaking the law can be perceived as leading to a superior societal outcome in the short run. This is the case mainly for two reasons. First, most of the negative consequences of corruption take some years to materialize or to be perceptible. Second, because politicians can justify their malfeasant behavior by claiming it to be in the best interests of the community. Nevertheless, most of the literature has ignored the fact that, in the short run, certain corrupt activities can be perceived or be justified by their capacity to improve the welfare of a community (cf. Fernández-Vázquez, Barberá, & Rivero, 2016).
In addition, prioritizing the outcomes over the integrity of a political decision might not be the only causal mechanism linking malfeasance to voting behavior. According to the retrospective voting perspective, voters should intrinsically value the provision of favorable outcomes and decide if they support the incumbent based on these considerations. This chapter, however, proposes that the causal mechanism that links malfeasance and vote choice is not only based on retrospective (sanctioning) considerations, but also on prospective considerations about the future (expected) conduct of candidates. I therefore expect that changes in character evaluations of politicians, which provide cues about the future behavior of political actors, are a relevant mechanism for explaining the impact (or the lack thereof) of malfeasance on voting choice. Previous empirical research has measured varying levels of tolerance for malfeasance in elections and has not yet been able to explain these differences
85 Competence corruption tradeoff
(for examples see Chong, De La O, Karlan, & Wantchekon, 2014; Ferraz & Finan, 2007). Understanding the causal mechanism that links information about malfeasance and the probability of voting for the politician in question is crucial because it provides the answer as to why tolerance of malfeasance might vary from case to case.
I test these arguments using an experiment that poses a dilemma; this allows us to assess the extent to which citizens prioritize procedural over outcome accountability or vice versa. Through this experiment the chapter assesses: (i) whether citizens infer different character traits depending on how politicians react to this dilemma; (ii) how these traits affect voting decisions; (iii) if the impact of procedural violations increases in cases of outright corruption. To do so, this chapter distinguishes between two degrees of malfeasance: breaking the law with and without obtaining a private benefit. Since only the latter can be considered corruption17, distinguishing between these two levels of malfeasance allows us to study citizens’ reactions to activities that, while entailing a procedural breach, cannot be considered corruption.
The results, based on an experiment fielded through a representative online survey of the Spanish adult population, indicate that voters punish malfeasant politicians, even if the politicians manage to achieve a superior outcome through their actions. Respondents prefer a law-abiding politician even when this implies obtaining a suboptimal outcome. The results show that this electoral punishment of malfeasance is primarily driven by the loss of trust provoked by the illegal decisions adopted in order to achieve an optimal outcome, which is not offset by gains in the perceived efficiency of the politician. This is especially apparent when the illegal decision involves a direct private gain for the politician.
These findings have relevant implications for the study of voting for corrupt politicians. By distinguishing between different degrees of malfeasance, the results indicate that citizens are only likely to tolerate procedural (legal) breaches that involve a side benefit for the population when politicians do not obtain any private benefit for themselves (i.e. when there is no corruption involved). In real
17 The most accepted definition of political corruption in empirical research is “the misuse of public office for private gain” (Nye, 1967). The private gain does not have to be an exclusive gain for the actor involved in the corrupt activity, it can also be a gain for the party the individual belongs to, family, friends, tribe etc. (Tanzi, 1998)
86 Chapter 4
settings where the short-term benefits of breaking the law are more tangible and the negative consequences are imperceptible, voters might be more willing to condone corruption. The analyses of the mechanisms driving this relationship clearly indicate that while side benefits might, in some cases, deter punishment for malfeasant politicians due to the perceived enhanced competence of these politicians, the loss in trust related to procedural breaches tends to offset these gains and leads to the punishment of politicians who break the rules.
4.2 Theoretical Framework:
4.2.1 Outcome accountability versus procedural accountability
To assess under what circumstances free and informed citizens support dishonest candidates, one must consider that while voters might care about integrity, they also consider other issues when deciding who to vote for. Corruption is only one of the many factors at stake for evaluating a candidate’s performance. The impact of a certain issue on an individual’s voting choice depends on the importance of this issue for the individual (Krosnick, 1988). Voters might implicitly trade politicians’ integrity for other characteristics that they value more (Rundquist et al., 1977).
Following this reasoning, recent research assesses whether citizens disregard malfeasance when candidates are capable of delivering other benefits, such as economic growth or other public goods. Observational evidence indicates that perceptions of corruption do not damage support for presidents in good economic contexts while, in bad economic contexts, they decrease it up to 10% (Zechmeister & Zizumbo-Colunga, 2013). These results are validated through survey experiments that show that citizens condone corruption in good economic contexts (Klašnja & Tucker, 2013), when the politician has a good management record (Muñoz et al., 2016), or when she implements favorable economic policies (Konstantinidis & Xezonakis, 2013). Moreover, a survey experiment fielded in Sweden, Spain and Brazil indicates that respondents would rather vote for a corrupt but competent politician than for an honest but
87 Competence corruption tradeoff
incompetent one (Esaiasson et al., 2014). However, the results of Chapter 3 only partly corroborate a tradeoff between the strong economic performance and the integrity of a candidate, when it is assessed in a multidimensional setting.
Recent observational research, in fact, indicates that when politicians cannot simultaneously maximize common welfare and behave honestly, citizens have a consistent preference for morally upright politicians (Allen et al., 2016; Allen & Birch, 2011; Birch & Allen, 2015). Allen et al., (2016) show that when asked to choose between the two characteristics, European citizens consistently prioritize politicians who are honest over politicians who deliver superior outcomes. However, at the same time, the analysis of Spanish municipal elections carried out by Fernández-Vázquez et al., (2016) shows that mayors' wrongdoings that generate side benefits for the community are not punished at the polls. These findings suggest that voters base their voting decisions on the outcomes generated by politicians’ decisions rather than on the procedural integrity of their actions. If, as a result of being corrupt, a mayor improves the aggregate welfare of a town, voters are likely to forgive her malfeasance. These findings clearly contrast with the opinions expressed by citizens in surveys, which reveal a clear preference for honest politicians, even if that implies that these politicians are less efficient at delivering public goods (see e.g. Allen et al., 2016). However, as discussed in previous chapters, standard survey questions could be biased by their failure to replicate real settings.
To address this concern, this chapter analyzes why and to what extent citizens might (or might not) support malfeasant politicians who manage to get the job done. However, unlike most previous studies, this chapter assesses a citizen’s likelihood of supporting politicians who manage to deliver superior outcomes precisely because they bypass established legal procedures or are openly corrupt. In this sense, this chapter comes closest to the aggregate-level analysis of Fernández-Vázquez et al. (2016), but allows us to disentangle the mechanisms that might drive the punishment of malfeasance at the polls or its lack thereof. I therefore test whether citizens favor outcome accountability over procedural accountability in situations that pit these two desirable properties against each other, i.e., circumstances in which politicians can either follow the established legal procedure and achieve a suboptimal outcome, or bypass that procedure in order to achieve an optimal outcome. I argue that presenting citizens with a situation that offers a direct tradeoff between these two desirable properties is
88 Chapter 4
the best way to reveal their priorities when it comes to procedural and outcome accountability.
Assessing citizens’ reactions to situations that pose a dilemma between outcome and procedural accountability is not only relevant in and of itself in order to evaluate theories of corruption voting, but also because these tradeoffs might actually be present when evaluating politicians before casting a vote. While malfeasance has a clear negative impact on society, this usually takes a long time to materialize and be perceived by citizens. For example, the negative consequences of building a school with lower quality materials than those budgeted for are not directly apparent, even though the eventual long-term outcome could be catastrophic. However, the positive impact of a new school in a community is unequivocal and directly perceptible in the short run (Fisman & Golden, 2017). When casting a vote, information about specific wrongdoings, if available at all, does not usually include details about the negative societal consequences of these actions. Therefore, it might be unfeasible to directly assess the negative impact of a specific act of misconduct when deciding who to vote for.
Furthermore, when politicians are caught red-handed, they sometimes justify their actions as being the best way to achieve an optimal societal outcome. Politicians have, for example, frequently justified the direct and discretionary assignment of a public contract to a company without holding a legally prescribed public tender because it was more efficient and cheaper (see for example Laura Cornejo, 2019; Raúl Rejón, 2018). Since the negative outcomes of misconduct are easily concealed, politicians can use these situations to justify their acts.
89 Competence corruption tradeoff
4.2.2 The mechanism: malfeasance, politicians’ traits, and voting decisions
The previous discussion implies that voters might intrinsically value the procedural integrity of politicians as well as the outcomes provided by them, and that this could directly shape their voting decisions. This argument goes in line with the retrospective voting literature. According to the retrospective perspective, we can expect voters to take decisions based on politicians’ past performance. This literature has mainly focused on the importance of economic performance for the vote (for a review see Lewis‐Beck & Stegmaier, 2007). The literature on corruption voting follows this perspective in order to understand why citizens vote for malfeasant politicians. A clear reflection of this is the fact that the literature talks about punishing malfeasant politicians, therefore taking a clearly retrospective position. Following this perspective, whether voters punish malfeasant behavior or not is indeed a matter of whether they prioritize outcome accountability over procedural accountability. However, while citizens do take politicians’ past performance into consideration, this might not be the only causal mechanism at play.
When studying voting behavior, the importance of candidates’ traits and evaluations of their characters has been extensively recognized (Campbell et al. 1960: 55; Bartels 2002; Goren 2007; Miller et al. 1986; Kinder 1986). These evaluation of politicians’ characters are an important determinant of voting choices because they provide relevant cues about the expected future behavior of candidates (Garzia, 2011). A limited and common set of relevant trait- dimensions that summarize the basis for political judgments and voting decisions have been identified across multiple contexts (Pancer et al., 1999, p. 362). These basic dimensions refer to the following traits of politicians: competence, integrity, and warmth/empathy (Funk, 1996). In this sense, the “good type” of politician that citizens should choose is frequently depicted as a principled, honest, skillful, competent and efficient person (Fearon, 1999: 59).
Opinions about politicians are largely based on the information citizens receive about these leaders’ behavior (Garzia, 2011; Fearon, 1999). People form their impressions of political leaders as they do for all sorts of individuals: by making inferences based on the information that is available (Funk, 1996). Politicians’
90 Chapter 4
observable behavior and decisions are consequential for the character traits that individuals associate with them. This should be the case in particular when politicians are confronted to solve a situation that poses a tradeoff or dilemma. Studies in social psychology indicate that reactions to situations that pose a dilemma between following one’s moral principles and achieving a superior outcome are highly consequential for the trait inferences that witnesses make about decision makers (Uhlmann et al., 2013).
I therefore expect that the way in which politicians react to a situation that poses a dilemma mold citizens’ evaluations of their characters and traits. Following a “selection model”, these character evaluations should, in turn, affect how likely it is that these citizens will vote for that politician. Figure 4.1 summarizes these arguments.
Figure 4.1. Analytical framework
This chapter focuses on two traits that voters value in politicians: their perceived trustworthiness (one of the key manifestations of integrity) and their perceived efficiency (a paramount manifestation of their competence). These traits capture the essence of the outcome vs. procedural integrity tradeoff, and therefore allow me to provide relevant insights about the reasons why citizens might (or might not) support corrupt politicians.
My basic expectation is that a politician who decides to follow established legal procedures, even when bypassing them would lead to a superior societal outcome, is inferred to be more trustworthy (for simplicity I refer to such a
91 Competence corruption tradeoff
politician as the legal politician). The study by Everett et al. (2016) on morality suggests that individuals who make these types of deontological judgments and decisions are, in fact, considered more trustworthy.18 Furthermore, following legal procedures provides a relevant cue about the reliability of the future behavior of a politician (Everett et al., 2016, Fearon, 1999). If a politician decides to follow the law, even when she has clear incentives not to do so, this is a fair signal that she will be guided to act by the same principles in future situations. This is relevant for the development of political trust, since individuals entrust power to authorities, but in many cases lack the knowledge or capacity to directly monitor the actions of these authorities (Levi & Stoker, 2000).
Following the same reasoning, I expect a politician who bypasses the law, even if she does so in order to achieve a superior societal outcome, to be considered less trustworthy. In this case, the behavior of the political authority provides a negative cue about the reliability of her future behavior. An authority who breaks the law once might be more likely to do so at other times, even that behavior would not benefit the common good. In fact, individuals who use this type of consequentialist reasoning tend to be associated with negative traits such as, for example, being selfish or self-interested (Kreps & Monin, 2014; Uhlmann et al., 2013).
Studies in social psychology indicate, though, that consequentialist actions can be considered more adequate when the negative consequences they entail are only a side-effect of the action (Cushman, 2013; Royzman & Baron, 2002), or when they are not directly intended by the actor (Waldmann & Dieterich, 2007). This leads me to expect that a politician who defies the law without obtaining a private benefit will seem more trustworthy than one who privately benefits from it, even if she also achieves an optimal societal outcome by doing so (for simplicity I refer to the former as the criminal politician and to the latter as the corrupt politician). If a politician obtains a private benefit as a result of breaking the law, one could doubt the motives that drove the politician to forgo the legal procedure. Since the political authority privately benefits from her action, one could think that breaking the law in order to obtain a private gain was the main
18 Deontology and consequentialism are two well-known perspectives on morality and ethics. While consequentialist theories focus on the maximization of happiness (or utility), deontological theories, emphasize the need to respect rights, duties and obligations (Everett et al., 2016).
92 Chapter 4
intention of the politician. This should provide an even more negative cue about the reliability of the corrupt politician’s future behavior, and should therefore reduce her trustworthiness further.
I therefore expect that:
H1.1: Changes in the perceived trustworthiness of politicians mediate the effects of corrupt, criminal and legal decisions on voting choice.
H1.2: Citizens place the greatest trust in legal politicians, less trust in criminal politicians, and the least trust in corrupt politicians (trust: legal > criminal > corrupt).
Focusing now on competence, I expect politicians who bypass the law in order to achieve an optimal societal outcome to be perceived as the most efficient ones. Bending some laws, when needed, should be perceived as belonging to the actions of a person that is more rational than a person who blindly follows legal procedures. The cost benefit calculation performed by pragmatic politicians who circumvent the established bureaucratic procedure to obtain a superior outcome signals that they are actors guided by a sense of rationality that dominates over other considerations such as the legality or morality of one’s actions. They are likely to adopt decisions that guarantee maximum societal welfare, avoiding the loss of utility that might be generated by following bureaucratic procedures. This should send a powerful cue about the efficiency that could be expected from these politicians, who could seem more competent at solving societies’ problems. Conversely, politicians who decide to follow the law, and as a result sacrifice the attainment of public goods, are considered less likely to weigh up costs and benefits in their decision-making processes, and more likely to be guided by a sense of morality and upright values. Consequently, these politicians are perceived as less efficient.
93 Competence corruption tradeoff
I therefore expect that:
H2.1: Changes in the perceived efficiency of politicians mediate the effects of corrupt, criminal and legal decisions on vote choice.
H2.2: Compared to legal politicians, the criminal and corrupt politicians will be perceived as the most efficient.
4.3 Empirical Strategy
The analyses draw on a vignette experiment included in an online survey. The survey was fielded in December 2018 through the online panel of the commercial firm Netquest. The sample of 1,200 respondents was recruited through quota sampling, including representative quotas for gender, age and education. The data were collected within the “LIMCOR: Limits to political corruption” project (Fundació La Caixa 2016 ACUPO177).
The design of the experiment was inspired by the sacrificial moral dilemmas commonly used in social psychology (Moore et al., 2008). The most common experiment in this field is the trolley dilemma, in which respondents are faced with a situation where they can either save a larger number of people by sacrificing a smaller number of people (consequentialist decision), or not sacrifice anybody but consequently allow a larger number of people to die (deontological decision). Through these experiments, researchers assess whether and why respondents prefer a deontological or a utilitarian solution. The experiment essentially mimics this design but is adapted to a more realistic situation that could occur in real life politics. The vignette describes a hypothetical situation in which heavy rain has flooded a town and has destroyed its sewage system, creating severe problems for the city residents (see Table 4.1). The law establishes that in order to repair the sewage system, the mayor of the town should hold a public bidding process (call for tenders) and assign the repair contract to the best bidder; however, this would delay the repairs. The mayor foresees that bypassing the legal procedure and directly assigning the repair to an experienced company would speed up the resolution of the problem, but doing so would be against the law.
94 Chapter 4
Table 4.1. Wording of the experimental vignettes19
Common introduction to the vignette:
Torrential rains have flooded and severely damaged the sewage system of a city. This situation generates severe problems for the city residents. In order to carry out repairs and solve the problem, the city council has to select a company through a public tender. However, the public call for tenders will delay the resolution of the problem. The only way to speed up the repairs is to avoid the call for tenders and directly assign the repair contract to an experienced company. However, this would not be legal.
Treatments:
Legal: David G. P/Laura G.P., mayor of the city and member of {respondent’s favorite party}, launches a public call for tenders that is finally awarded to an experienced company. This decision respects the established procedure, although it delays the repairs. Criminal: David G. P/Laura G.P., mayor of the city and member of {respondent’s favorite party}, does not launch a public call for tenders, and directly awards the repairs to an experienced company. This decision speeds up the repairs, although it does not respect the established procedure. Corrupt: David G. P. /Laura G.P, mayor of the city and member of {respondent’s favorite party}, does not launch a public call for tenders and directly awards the repairs to an experienced company that has contributed to the electoral campaign of his/her party. This decision speeds up the repairs, but it does not respect the established procedure.
After introducing participants to the situation, respondents were randomly assigned to one of the three different treatments. In the first treatment, the mayor follows the law but, as a consequence, there is a delay in solving the problem (legal decision). In the second treatment, the mayor sidesteps the legal procedure but prioritizes the outcome (criminal decision). Finally, in the third treatment, the mayor also prioritizes the outcome and does not follow the legal procedure but, at the same time, this decision involves a private gain for the mayor’s political party (corrupt decision). This third treatment therefore closely matches most definitions of corruption. While in the second treatment the
19 The original wording of the experiment (in Spanish) can be found in the Appendix C Table C.1.
95 Competence corruption tradeoff
mayor just violates the law, in the third treatment the mayor violates the law and potentially obtains a private gain. Hence, the degree of malfeasance is higher in the third treatment condition.20 The corrupt activity described in this experiment is a wrongdoing commonly seen in Spanish politics, where politicians do not hold a public tender, but assign a public contract to a company that gives illegal donations to their political party (Jareño Leal, 2017). This kind of practice has become common knowledge among Spanish citizens due to several recent corruption scandals such as the Gürtel scandal or the 3% case.
Several reasons justify the specific design of the vignettes. First, confronting participants with a direct tradeoff where the action can either maximize the outcome by violating the procedure established by the law or can respect the procedure and attain a suboptimal outcome is the best way to understand the true preferences of respondents. Since procedural integrity, the absence of corruption, and competence in office are highly valued characteristics, the best way to assess which of them respondents truly consider as more important is to present them with a direct tradeoff (see Edmons, 2014 for a review of similar arguments applied to moral dilemmas). Second, the situation presented to respondents is due to a natural disaster; in this way, they cannot blame the mayor for her bad management in preventing the problem she must now resolve. By linking the problem with heavy rains, the mayor is only responsible for the way she addresses the problem, while minimizing her responsibility in the actual generation of the problem. This design maximizes the internal validity of the experiment as the likelihood of supporting the mayor and the inferences about her traits can only be traced back to the decisions she adopts. Third, the situation in which a public administration does not assign a contract through a public bidding process or a call for tenders is highly realistic, as this is one of the mechanisms where we most commonly find corruption in Spanish politics (Jareño Leal, 2017) 21. Therefore, the situation presented is highly realistic and
20 A manipulation check (discussed in detail in the Appendix C, Section C.1) confirms that the manipulation altered the perceived malfeasance. When asked about the extent to which they think that the mayor described in the vignette is corrupt, respondents in the third treatment group are much more likely to think that the mayor is corrupt. 21 These are two recent examples of similar illegal practices in public bidding processes in Spain. https://www.eldiario.es/sociedad/Gobierno-troceo-presupuesto-adjudicar- contratos_0_772373776.html https://www.eldiario.es/cyl/AUDIO-Fomento-Castila-Leon- Enredadera_0_874213000.html
96 Chapter 4
respondents should be well used to reading about this kind of violation of the law and the dilemmas these situations entail.
Having a treatment with different degrees of malfeasance allows me to disentangle the effect of corruption and to determine whether citizens are mainly concerned with the legality of the procedure or with the fact that the wrongdoing entails a private gain. Furthermore, it allows us to assess a respondent’s reaction to a procedural breach that is less stereotyped as “corruption”. This is relevant because, as discussed earlier, corruption is an issue that Spanish citizens express being highly concerned about. Therefore, respondents in the survey experiment might have strong opinions that reject any form of corruption. By presenting a violation of the law that is not linked to corruption, I am able to gauge respondents’ true preferences about procedural integrity. Differentiating between an illegal decision and a corrupt one also lets us assess if citizens infer distinct personal traits of a politician depending on whether she obtains a private gain or not.
In the vignette presented to respondents, I control away the effect of partisanship, as previous studies have shown how partisanship might bias perceptions about corruption (Anduiza et al., 2013; Solaz et al., 2018). The vignette always refers to a mayor of the respondent’s preferred party, so that the effects are comparable across all respondents. Respondents who declare no partisanship are excluded from the main analyses discussed below. Controlling for partisanship is fundamental for three reasons. First, when deciding who to vote for, partisanship is a cue that is always available for voters. Second, if I presented a fictitious mayor with no party affiliation, respondents might infer her partisanship from the mayor’s behavior. Third, adding real party labels increases the realistic nature of the vignette: in real life, voters always know the party a politician belongs to.
Partisanship is measured through a pre-treatment question that asks respondents if there is any party that they feel more sympathy towards or that they think is closer to their ideas. If they answer that they do not feel close to any of the parties, they were asked a follow-up question that asked what party they would select if they had to choose one. Even so, 19% of the respondents declared that they did not feel close to any of the Spanish parties. These respondents were randomly assigned to one of the four parties with more representation in the
97 Competence corruption tradeoff
Spanish parliament (PP, PSOE, Podemos and Ciudadanos) at the time the experiment was conducted. In addition, they were also excluded from the main analyses discussed below. Including these respondents in the analyses does not alter the conclusions, however (the results are available in Appendix C section C.4).
The gender of the mayor was also randomized to control that respondents wouldn’t infer the gender of the politician form other information.
The main dependent variable of the analyses is the propensity to vote for the mayor. This variable is measured with a direct question asking respondents how likely it is that they would vote the mayor they have read about in the vignette, with 0 being “would never vote for this mayor” and 10 “would certainly vote for this mayor”.
To analyze the hypothesized causal mechanisms, I asked respondents to evaluate if the mayor described in the vignette possessed certain traits. This question draws on the format commonly used in the American National Election Study to enquire about candidates’ traits: Next, you will see a series of expressions that people can use to describe a politician. To what extent does each expression describe the mayor of the text you just read? “he/she is efficient”, “he/she can be trusted”, “he/she cares about the needs of people like me”. For each of the statements, respondents could either answer quite well or quite badly. These questions allow us to measure the three basic trait dimensions discussed above: competence (efficiency), integrity (trustworthiness) and empathy (caring about others). I am thus focusing on the three most important traits according to the literature.
98 Chapter 4
4.4 Results
4.4.1 Average treatment effects
Figure 4.2 summarizes the propensity to vote for the mayor depicted in the vignette among respondents assigned to each of treatment groups. The average propensity to support the mayor, who is always from each respondent’s party, is 5.58 on a 0-10 scale (where 10 indicates that the respondent is completely sure that she would vote for that mayor). This propensity to vote for the mayor is significantly different across the three experimental groups. The mayor that obtains the greatest average support is the one that follows the laws, even if that leads to a suboptimal outcome (7.44). The mayor that bypasses the law to achieve a superior outcome (i.e. assigns the repair to an experienced company without holding a public tender in order to solve the problem faster) gets an average support of 6.29 (criminal decision). Finally, the corrupt mayor, who directly assigns the repair to a company that solves the problem faster but has contributed to her party campaign, receives an average support of 4.37 (corrupt decision). The support for this mayor is, therefore, 3.07 points lower than the support enjoyed by the legal mayor.
Figure 4.2. Mean propensity to vote across different treatment conditions
10
9
8
7
6
5
4
Meanpropensity to vote 3
2
1
0
Legal decision Criminal decision Corrupt decision
Note: 95% confidence intervals around the means
99 Competence corruption tradeoff
These results indicate that, on average, respondents prefer a mayor who respects the integrity of the procedure if that implies achieving a suboptimal outcome, rather than one that achieves an optimal outcome by bypassing the legal procedure. It is worth noting that while the legal and criminal mayors are more likely to be supported than not (their average support is higher than the scale midpoint: 5), the corrupt mayor has an average support that is lower than the scale midpoint. Therefore, respondents are not likely to support a mayor from the party they identify with if the politician obtains a superior societal outcome by corrupt means. Hence, respondents are much more likely to defect from their own party when violations of procedural integrity involve a private benefit.
Figure 4.3. Analyses of traits: proportion of respondents that consider the mayor trustworthy and efficient across different treatments conditions
1
.8
.6
Proportion .4
.2
0 Politician is trustworthy Politician is efficient
Legal decision Criminal decision Corrupt decision
Note: 95% confidence intervals around the means
100 Chapter 4
Focusing on the personal traits that respondents infer about politicians that prioritize either superior outcomes or procedural integrity, Figure 4.3 summarizes the proportion of respondents believing that each of the mayors depicted in the vignette can be associated with a given trait. The mayor who follows the law clearly inspires the greatest trust, the criminal mayor follows her, and the one that is considered the least trustworthy is the corrupt mayor. It is worth noting that not only the vast majority of respondents (around 80 percent) trust the mayor who respects procedural integrity, but also that a significant number of respondents (almost 60 percent) believe that the criminal mayor is someone who can be trusted. This is completely different for the mayor who openly engages in corruption, since in this case only 39 percent consider her trustworthy. Therefore, in line with my expectations, the degree of malfeasance has a clear negative impact on politicians’ trustworthiness.
Turning now to the perceived efficiency of the mayor, the politician who bypasses the established legal procedures in order to obtain a superior societal outcome, but at the same time does not obtain any private gain is considered the most efficient of the three. Both the mayor that follows the legal procedures and the mayor that engages in corruption are considered less efficient. These results are only partially in line with my expectations. Since both the criminal and the corrupt mayors prioritize outcomes and, in fact, achieve a superior outcome, I would expect them to be considered equally efficient. However, the corrupt mayor obtains the same evaluation regarding her efficiency as the legal mayor, although the former achieves an optimal outcome while the latter does not. One might speculate that individuals might doubt the motives that lead the corrupt mayor to opt for the course of action that achieves the optimal outcome if that decision also brings her private benefits. In that case, the inference about the present and future efficiency of that mayor might be less certain, since one does not know if she would also prioritize the achievement of optimal societal outcomes if she has nothing to gain from that decision. This will be tested at the end of the next section.
101 Competence corruption tradeoff
4.4.2 Mediation analysis
To test the hypothesized causal mechanisms at work, I now turn to mediation analysis. Traditional mediation analysis implicitly assumes the independence of the competing causal mechanism (Imai et al., 2011). According to this assumption, it would be presumed that the suggested mediators (trustworthiness and efficiency) have no influence on each other. In this case, the independence assumption seems too restrictive on theoretical grounds. The inferred efficiency of the politician could have an impact on her perceived trustworthiness, and vice versa. Consequently, I turn to mediation analysis for multiple causally dependent mechanisms following the framework proposed by Imai and Yamamoto (2013), which is implemented through the multimed function in R developed by Tingley, Yamamoto, Hirose, Keele and Imai (2014). By using this method, the assumption of independence of the mediator is relaxed. Moreover, within this framework one can test the sensitivity of the results to violations of the key identifying assumptions: sequential ignorability22 and homogeneous interaction23. To increase the plausibility of the sequential ignorability assumption, all the mediation models include the following pre-treatment covariates: sex, age, education, political interest, social trust, ideology, ideology squared, and a variable measuring the strength of each respondent’s party identification. As such, I address the main criticism against mediation analysis – that it does not independently manipulate the mediators –, by controlling for both (i) possible covariates that could be correlated with the mediator and the independent variable, and for (ii) another possible mediator that could influence the main mediator of interest (see Green, Ha, and Bullock 2010; Bullock, Green, and Ha 2010). In addition, following the recommendation of Imai and Yamamoto (2013), I also assess the sensitivity of the findings against violations of the sequential ignorability assumptions using the test summarized in Figure C.3 in Appendix C.
First, I estimate the average causal mediation effect (ACME), which represents the causal effect of each of the treatments on the likelihood of voting for the mayor that are transmitted through the mediator of interest. Second, I estimate
22 The main mediator is assumed to be exogenous after controlling for the alternative mediator, the treatment and the pretreatment confounders (for a detailed explanation see Imai & Yamamoto, 2013b). 23 The model assumes no interaction between the two mediators.
102 Chapter 4
the average direct effect (ADE), that represents the rest of the treatment effect that is transmitted either through the alternative mediator or through other mechanisms. Finally, the total effect is obtained by adding the ACME and the ADE, and represents the full change in the outcome variable that is attributable to the treatment.
I first focus on the models which treat trustworthiness as the main mediator and efficiency as the alternative mediator. These models estimate the effects of the treatments on the likelihood of voting for the mayor that are mediated by changes in the perceived trustworthiness of the politician, while taking into account that there might be an additional character trait at play –such as efficiency– that is related to trustworthiness and that might also mediate the effects of the treatments.
Figure 4.4 shows that the relationship between the treatment and the outcome is significantly mediated by changes in the perceived trustworthiness of the mayor. 44 percent of the effect of the mayor that defies the established legal procedure on the probability to vote is mediated by the decrease in trust (taking the legal mayor as the control baseline). Similarly, when comparing the corrupt and the criminal mayors, a further decrease in trustworthiness accounts for 38 percent of the reduction in the likelihood of supporting the corrupt mayor (compared to the criminal mayor). These results indicate that the drop in trustworthiness is one of the prime factors that explains why the legal mayor tends to be the one preferred. The violations of procedural integrity, especially if they also involve a private gain, generate a substantial drop in the belief that a politician is trustworthy, and, in turn, this belief has a sizable negative effect on the likelihood of supporting a politician. This result is in line with the idea advanced in recent accounts of the relationships between politicians’ traits and electoral behavior that indicate that integrity-related traits are of prime relevance for voting decisions (Laustsen & Bor, 2017).
103 Competence corruption tradeoff
Figure 4.4. Mediation analysis of trustworthiness with confounding alternative mechanism (efficiency)
ACME
ADE
Total Effect Criminal vs Legal Corrupt vs Criminal Corrupt vs Legal
-4 -3 -2 -1 0 Effects on probability to vote Percent mediated by trust: Legal vs Criminal 44.45% / Criminal vs Corrupt 37.55% / Legal vs Corrupt 36.53%
The second sets of models treat efficiency as the main mediator and trustworthiness as the alternative mediator. These models estimate the effects of the treatments on the likelihood of voting for the mayor that are mediated by changes in the perceived efficiency of the politician.
Figure 4.5 indicates that the relationship between the treatment and the outcome is effectively mediated by efficiency. However, the extent to which efficiency mediates the treatment effects is substantially smaller when compared to the mediation effects of trustworthiness. Gains in the perceived efficiency of the politician only mediate the negative effect of adopting an illegal decision on the likelihood of supporting the mayor by 11 percent (when compared to the mayor who adopts a legal decision). This means that efficiency mitigates the drop in the likelihood to vote for the mayor, as respondents consider the criminal mayor more efficient than the legal mayor. However, while gains in efficiency seem to reduce the electoral punishment attributable to the decision adopted by this mayor, they are not enough to completely offset the negative effects of her decision. The perceived loss in efficiency of the corrupt mayor, with respect to the criminal mayor, contributes to exacerbating the electoral punishment of the
104 Chapter 4
mayor that obtains a private gain through her decision. In this case, changes in efficiency mediate the effects of adopting a corrupt decision on the likelihood to vote for that mayor by 14 percent (in comparison to the likelihood of supporting a politician who adopts a decision that bypasses the law but does not involve a private gain). Therefore, it appears that in the case of the corrupt mayor, a loss on both perceived efficiency and trustworthiness contribute to the stark electoral punishment of this politician.
Figure 4.5. Mediation analysis of efficiency with confounding alternative mechanism (trustworthiness)
ACME
ADE
Total Effect Criminal vs Legal Corrupt vs Criminal Corrupt vs Legal
-4 -3 -2 -1 0 Effects on probability to vote Percent mediated by efficiency: Legal vs Criminal -11.17% / Criminal vs Corrupt 13.51% / Legal vs Corrupt 2%
I expected that in a comparison between a politician who prioritizes outcomes and one that prioritizes the integrity of the procedure, the former would always increase her electoral support through perceived gains in efficiency. However, this does not seem to always be the case. The comparison between the legal and the corrupt mayor indicates that efficiency does nothing to deter the punishment of the corrupt mayor since, in this case, the ACME value is almost 0. As I argue above, it is possible that citizens might doubt the reasons that motivate the corrupt mayor’s decision. This might weaken the inferences voters make about the efficiency of that politician. In order to test this argument, I estimate a set of
105 Competence corruption tradeoff
models with empathy as the main mediator and efficiency as the alternative mediator24. Whether respondents consider a politician empathic or not will surely be correlated with the reasons they think are driving the politicians’ actions. If the corrupt politician is perceived to be guided by his own interests rather than by the will to increase the community’s wellbeing (outcomes), they should consider the corrupt politician to be less empathic than the criminal politician. Figure 4.6 shows that empathy does indeed significantly mediate the relationship between the politician’s behavior and the propensity to vote for her. The perceived loss in empathy of the corrupt mayor, with respect to the criminal mayor, increases the electoral punishment of the mayor that obtains a private gain thanks to wrongdoing.
Figure 4.6. Mediation analysis of empathy with confounding alternative mechanism (efficiency)
ACME
ADE
Total Effect Criminal vs Legal Corrupt vs Criminal Corrupt vs Legal
-4 -3 -2 -1 0 Effects on probability to vote Percent mediated by empathy: Criminal vs Legal -13.76% / Corrupt vs Criminal 32.09% / Corrupt vs Legal 9.55%
24 The results displayed here are based on the model that uses efficiency as an alternative mediator. See Figure C.2 in Appendix C for the results of the models with the alternative mediator of trust.
106 Chapter 4
As a final consideration regarding mediation effects, the sensitivity analyses summarized in Figure C.3 in Appendix C indicate that the ACME estimates for efficiency are highly sensitive to violations of the homogeneous interaction assumption. Therefore, the results from this experiment only provide weak support for the expectation about the mediating effects of inferring the efficiency of a politician who adopts a decision that prioritizes either achieving superior outcomes or respecting procedural integrity. Conversely, the sensitivity analyses reveal that the conclusions are far more robust when it comes to the mediating effects of trustworthiness.
4.5 Conclusions
The experiment implemented in this study allowed me to assess the causal effect of a political decision that achieves an optimal outcome but violates procedural integrity. The results clearly show that politicians’ different responses to situations that pose a dilemma between achieving a superior outcome or respecting the legally established procedure affect citizens’ likelihood of supporting them.
Respondents prefer a politician who follows the rule of law even when this implies achieving a suboptimal outcome. However, it is remarkable that the politician that breaks the law to obtain an optimal outcome without obtaining a private benefit from it is more likely to be supported than not (her average support is higher than the midpoint on the scale). Yet, the mean support of the politician drops sharply when she defies the law to achieve an optimal outcome and, at the same time, obtains a private gain from it. When politicians violate the law to achieve an optimal outcome and they do not privately benefit from it, citizens are much more likely to support them than when their decision involves a direct private gain. According to the results, voters are more likely to tolerate wrongdoing when it serves to improve the conditions of their community. However, this is only the case when they cannot directly perceive that the politicians have privately benefited from breaking the law.
107 Competence corruption tradeoff
The results also confirm that respondents draw inferences about the traits of politicians depending on the decisions they adopt. The causal relationship between malfeasance and the drop in the likelihood of supporting the mayor is mainly caused by the erosion in the perceived trustworthiness of the mayor. Although respondents do consider the mayor that defies the established procedure to obtain an optimal outcome as more efficient, in this experiment, efficiency only slightly mitigates the drop in support caused by the loss of trust. The gains in the perceived efficiency of a mayor that breaks the law in order to obtain a desirable outcome for the community are not enough to overcome the decrease in trust. To sum up, trustworthiness is the main causal mechanism that drives the reduction in the likelihood of voting for a corrupt politician. These results reconcile the literature on the attitudinal effects of corruption, which finds a strong negative effect of corruption on trust, and the literature on the behavioral effects of corruption.
Citizens’ forceful rejection of corruption could limit the ability of experiments to measure the support of politicians that defy legal procedures. This chapter addresses this limitation by including a treatment of a procedural breach that is not directly linked to corruption. Thanks to this gradual treatment of malfeasance, this chapter is better suited for gauging respondents’ true preferences about procedural integrity. The gradual treatment of malfeasance allows us to better understand why in real settings respondents might be more willing to condone corruption, since in certain situations, the short-term benefits of breaking the law are more tangible and the negative consequences might be imperceptible.
108
Chapter 5
Who cares? Individual heterogeneity in the punishment of corruption
5.1 Introduction
Thus far, this dissertation has assessed different interpretations that might explain the rather lenient punishment of corrupt politicians. I have provided evidence that generally, citizens do truly care about corruption, but that in multidimensional settings, such as elections, they also consider a politician’s other characteristics. Co-partisanship and − under certain conditions − strong economic performance, can moderate the electoral punishment of corruption. While these results apply to all voters on average, the relationship between corruption and the vote may be unequal across voters. To draw an even more precise picture of attitudes towards corruption and their political consequences, this chapter focuses on individual predictors of the relative importance of corruption on voting choice. Identifying the determinants that correlate with a higher probability of trading off corruption at the individual level provides, us with a more accurate understanding of why dishonest politicians are punished only very leniently in elections.
109 Individual heterogeneity
Research on the individual predictors of attitudes towards corruption is scarce. Some studies have identified age, gender, employment status and income as determinants of attitudes towards malfeasance (Anduiza et al., 2013; Gatti et al., 2003; Riera et al., 2013; B Torgler & Valev, 2006). Nevertheless, the extant literature bases its analysis on citizens’ individual differences in condemning a certain corrupt activity (cf. Riera et al. 2013). While these studies might be interesting for finding out citizens’ opinions regarding these types of wrongdoings, as discussed in Chapter 2, respondents’ statements in surveys might differ greatly from how corruption scandals actually shape their vote. When asked in the abstract context of a survey question, respondents may truly consider corruption as an essential issue that determines their vote, and express that. However, when confronted with the tradeoffs they face in elections, the weight that corruption has on their vote decreases substantially. Therefore, this chapter assesses the correlation between citizens’ individual attributes and the relative importance they give to corruption when casting a vote in a multidimensional scenario. Inspired by the results of previous work, this chapter specifically focuses on gender, political sophistication and age.
Since the publication of a study that identified higher female representation in politics with lower corruption, many have been searching for explanations to understand this relationship (Dollar et al., 2001; Swamy et al., 2001). These explanations can be grouped into two main theories. The first posits that women and men differ emotionally and psychologically. The argument is that women simply have higher ethical standards than men (Benno Torgler & Valev, 2010). However, the second theory argues that women are only less corrupt than men due to a function of opportunities and constraints (Stensöta and Wängnerud 2018), since currently women have fewer important public positions and less access to the networks where the corrupt activities are taking place. These two theories have different implications for whether women or men care less about a politician’s corruption. If women have higher ethical standards than men, we would expect them to punish corrupt politicians more. However, if women are simply less corrupt due to a function of lower opportunities, then we would expect them to react to corruption to the same degree as men.
As far as political sophistication is concerned, there are two different arguments regarding the relationship between sophistication and corruption accountability. One argument indicates that more sophisticated individuals should be able to
110 Chapter 5
access more and better political information (Riera et al., 2013). As a result, they should be more aware about a politician’s wrongdoings and their negative consequences for society. According to this argument, sophisticated individuals should punish corrupt politicians more harshly than less sophisticated individuals. However, insights from research on motivated reasoning also show that sophisticated individuals have the motivation and skills to refuse messages that go against their previously held attitudes (Zaller 1992). Consequently, sophistication, especially when it is in interaction with partisanship, could undermine corruption accountability.
The relationship between corruption and voting choice may also vary across voters’ ages. This could be due to two different causes: cohort differences in tolerating corruption, or life cycle effects. Variations among cohorts arise when generations experience differing socialization processes (Wagner & Kritzinger, 2012). Different historical events can influence the political views of individuals when they become politically aware, and shape their future political behavior. Conversely, life cycle effects refers to the value changes that occur with ageing. As individuals grow older, their needs and interests may change and this can have an impact on their political behavior (Braungart & Braungart, 1986). The implication of these two theories on corruption accountability are discussed in the next section.
The “Theoretical framework” section develops the different arguments outlined by the literature on gender, political sophistication and age, and their implication for corruption accountability. After that, I present the empirical strategy of this chapter based on an original survey experiment implemented for this dissertation. The “Results section” shows that although the results – that voters care about corruption but in multidimensional decisions they are willing to trade off integrity – are in fact universal among different types of people, there is some variation across the sophistication and age of individuals. In the last section I discuss the implication of these findings.
111 Individual heterogeneity
5.2 Theoretical framework
5.2.1 Gender
An increasing interest in the relationship between gender and corruption has emerged since the publication of two studies that linked more female representation in politics with less corruption (Dollar et al., 2001; Swamy et al., 2001). These finding led to a set of research pieces that tried to unravel the nuances of this relationship. Nevertheless, there is still no consensus regarding the mechanism behind this pattern (Eggers et al., 2018). The many arguments that have been put forward for understanding the relationship between women in parliament and lower levels of corruption can be grouped into two main theories. The first refers to important cognitive and emotional gender differences to explain why women have a lower degree of criminality; these could be an expression of different “biological, psychological and experiential realities” (Benno Torgler and Valev 2010:554). The second theory explains gender differences by alluding to the different constraints and opportunities that women and men face in public life. The following paragraphs describe the arguments used in each of these theories.
Women are altogether less likely to engage in crimes or violent behavior than men (Kruttschnitt, 1994). Some scholars argue that this is the case because women have higher ethical standards. They explain lower levels of female criminality by citing women’s higher levels of empathy and pro-social behavior, which lead them to be more aware of the repercussions of their acts on others (Broidy et al., 2003; Mestre et al., 2009). A slightly different argument, but one that also highlights psychological differences, is the one that posits that women are more averse to risk-taking than men (Esarey & Chirillo, 2013; Esarey & Schwindt-Bayer, 2018). According to this argument, women behave better in office not because they have higher ethical standards, but simply because they are more afraid of the punishment they could face.
Regarding the punishment that politicians face, another argument posits that women behave better in office than men because of voters’ expectations of female politicians. According to this argument, in public life, women face higher standards. It is these higher levels of accountability that force them to behave
112 Chapter 5
better in office, not their actual higher ethical standards (Esarey & Schwindt- Bayer, 2018). A recent study actually shows that on average, female politicians do not suffer harsher punishment for their misconduct (Eggers et al., 2018). But even if this is not actually the case, a belief that they will be more harshly punished could already be enough to condition their behavior in office.
However, an alternative strand of the literature argues that the relationship between female representation in government and national levels of corruption has been misinterpreted. According to this argument, the initial reading of this correlation is a typical case of reverse causality. Scholars have claimed that the presence of women has led to lower levels of corruption, but corruption could also work as a hurdle for the political recruitment of women. The suggestion is that the shady networks that are required for corruption to survive have rigid access barriers for newcomers, such as female politicians (Sundström & Wängnerud, 2014). According to this argument, women participate less in corruption simply because they have more obstacles to surmount to reach the corrupt networks, not due to their higher ethical standards or greater risk aversion. A related argument is one that posits that the correlation between female representation and corruption is simply due to a function of opportunities (Stensöta and Wängnerud 2018). Women are currently less involved in corrupt activities because they have altogether a smaller role in positions where those behaviors are prevalent.
While this is a fascinating research agenda, this chapter is concerned with something else: how women and men react to corrupt politicians – rather than who is more or less involved in these activities. Research has so far identified gender differences in justifying corrupt activities. Nevertheless, the direction of this relationship provides mixed evidence. Some studies find that women are harsher in condemning tax evasion and those accepting bribes than men (Swamy et al., 2001; Benno Torgler & Valev, 2010). Although another study finds exactly the opposite, women express less negative views of a mayor that favored family members (Anduiza et al., 2013).
This chapter specifically addresses the relationship between gender and the relative importance of corruption when casting a vote. It therefore assesses whether political corruption has a different effect on voting choice for men and women. So far, the research that has assessed this seems to provide evidence
113 Individual heterogeneity
that women punish corruption more harshly than men. A survey experiment run in Great Britain finds that female respondents punish malfeasance more severely (Eggers et al., 2018). However, a study that combined survey and election results in Spain only finds rather weak evidence for this hypothesis (Riera et al., 2013). Since further research is clearly required, this chapter provides an additional test for assessing gender differences in corruption accountability.
While the theoretical discussion above has focused on the differing levels of criminality between genders, the arguments represented lead to different implications for the issue of interest in this chapter. If women have higher ethical standards and are more pro-socially driven than men, it is intuitive to expect that they also condemn corruption more severely and therefore punish corrupt politicians more. However, if women simply participate less in corruption because of the constraints they face in public life, then there should be no gender differences in the electoral punishment of corruption, since both genders should value integrity equally.
As far as the tradeoff of integrity against co-partisanship is concerned, Chapter 3 has shown that on average, partisanship is equally important in determining voting choice and that co-partisanship moderates the negative effects of corruption. Considering that women are on average less politically engaged (Verba et al., 1997), one could also expect them to have softer partisan ties. Consequently, a politician’s partisan attachments might moderate the negative effect of corruption on the vote less strongly for women than for men.
5.2.2 Political sophistication
As far as political sophistication is concerned, there are two different arguments on how sophistication could make its impact on electoral accountability for corruption. One states that more sophisticated individuals should be better informed about a politician’s wrongdoings and more aware of the negative consequences of corruption for society. Therefore, politically sophisticated individuals would be expected to punish corruption more severely than less sophisticated individuals. However, insights from research on motivated reasoning also shows that sophisticated individuals have more abilities and strategies to avoid the cognitive dissonance between liking a political party or
114 Chapter 5
politician and finding out about their wrongdoings. Consequently, sophistication in interaction with partisanship could undermine corruption accountability. The next two paragraphs develop these arguments.
Not all individuals are equally exposed to political information (Riera et al., 2013). An individual’s political interest and knowledge is likely to have an impact on how informed they are about a politician’s wrongdoings. Even if a scandal is widely talked about in the news and all citizens are exposed to the information, it is still a complex task to understand corruption charges, who is responsible for them and their negative consequences for society. Therefore, it is likely that sophisticated individuals are more aware of the implication of political wrongdoings and to know who to blame for them. A set of empirical studies support this argument. Evidence shows that sophisticated citizens, when faced with information about corruption, discern more easily between credible and less credible information (Weitz-Shapiro & Winters, 2016). In addition, politically-aware individuals are more likely to rely on issue-based considerations than on heuristics to form their opinions on new issues (Kam, 2005). Consequently, when confronted with information about wrongdoings, politically-aware citizens should be more likely to identify the actual content of the message, rather than be biased by the party the wrongdoing is linked to. In line with this argument, Anduiza, Gallego, and Muñoz (2013) find that politically knowledgeable individuals judge a mayor involved in a corruption scandal more harshly, and that political knowledge reduces the moderating effect of co- partisanship.
However, insights from the research on motivated reasoning point to a different effect of political sophistication on punishment for corruption at the polls. Sophisticated individuals have more abilities and strategies for avoiding the cognitive dissonance between liking a political party or politician and finding out about their wrongdoings. In this way, sophisticated voters have the motivation and skills to refuse messages that go against their previously held attitudes (Zaller 1992). They also have more abilities to seek out information in favor of their preexisting attitudes (confirmation bias) and to use reasoned standpoints to argue against evidence that contradicts their previously held attitudes (disconfirmation bias) (Wagner et al., 2014). Furthermore, sophisticated individuals should be more motivated to hold onto their opinions because they invested more time in formulating them (Lodge & Taber, 2012). According to
115 Individual heterogeneity
this discussion, sophisticated individuals should be more likely to condone corruption in their preferred political party. In line with this argument, Wagner, Tarlov, and Vivyan (2014) find that the moderating effect of partisanship on the evaluation of political misconduct is greater among voters with higher levels of political sophistication.
Two expectations, which apparently seem contradictory but that could coexist, are derived from the discussion of these two arguments. First, based on the theory that posits that sophistication leads to greater awareness, corruption should have a stronger negative effect on the probability of voting for a politician among sophisticated individuals than among less sophisticated individuals. Second, insights from research on motivated reasoning lead to the expectation that the moderating effect of co-partisanship on the negative effect of corruption is stronger among politically sophisticated individuals.
The first expectation has been tested in a study that combined survey data and election results, but the results did not show a significant difference for the political sophistication of respondents (Riera et al., 2013). As discussed in the empirical section, the methodological strategy of this chapter differs greatly from the Riera et al. (2013) study. Consequently, this chapter is a further test for the relationship between political sophistication and corruption accountability. Furthermore, in this chapter, respondents with different levels of sophistication are all equally likely to be exposed to the information about corruption. So, if we do find significant effects, it is only attributable to how respondents interpret the received information, and not to whether they are exposed to it or not. As far as the second expectation is concerned, in this chapter I can test whether or not the moderating effect of partisanship on corruption varies across different levels of political sophistication.
116 Chapter 5
5.2.3 Age
The relationship between corruption and voting choice may also depend on a voter’s age. Age is a primary social category of human existence that is used in societies to assign roles and grant power (Braungart & Braungart, 1986). The impact of corruption could vary among different age groups due to cohort differences in tolerating corruption or due to life cycle effects.
Variations among cohorts arise when generations experience differing socialization processes. This happens because the values and issues that are being discussed during the years when citizens become politically aware, and that shape their future political behavior, can change over time (Wagner & Kritzinger, 2012). The argument is that important historical events – such as wars, economic crises, significant technological shifts or cultural changes – have an especially strong influence on the political attitudes of individuals in their formative stages. These events shape their political views and, in this way, members of different cohorts can develop different values and political preferences. A key assumption of the cohort theory is that attitudes and political behavior develop during youth, rest constant over time and impact future political behavior (Braungart & Braungart, 1986). If differences in political behavior among age groups are due to a cohort effect, then we can expect a generational replacement. At some point in time, the political behavior of young voters will become the mainstream behavior.
On the other hand, life cycle refers to the effect of ageing on voters’ political behavior. The argument followed by this approach is that as people grow older they go through certain changes in “physiology, cognitive functioning, emotional patterns and needs” (Braungart and Braungart 1986:208). Each stage in life is related to different interests and needs and this can of course impact one’s political preferences. Changes in status such as getting married, losing or changing one’s job or having children could influence people’s values. Therefore, as voters grow old, they may vary in what they consider important when casting a vote. According to this argument, different stages in the life of voters could explain the variance among voters’ tolerance towards corrupt politicians. If this is the case, we would not observe a generational replacement and the average importance of corruption would only vary due to demographic changes in the population.
117 Individual heterogeneity
Due to the lack of panel data, in this chapter I can only assess whether the impact of corruption on voting choice varies across different age groups, without establishing whether this is due to cohort or life cycle effects. Nevertheless, these two explanations could be pointing in different directions regarding the relationship between age and tolerance towards corruption.
Making sense of the fact that in the last two decades there has been an increase in citizens’ concern regarding corruption (Fisman & Golden, 2017; Villoria Mendieta & Jiménez Sánchez, 2012), one could think that this is so due to a generational replacement. While older generations were more indifferent to political corruption, younger generations are more averse to it. This could be the case in Spain particularly, where the older generations were socialized during the dictatorship or in the period of transition to democracy while the younger generations were socialized in an established democracy. The older generations might therefore have been less familiarized with discussions on quality of government and more preoccupied in overcoming the dictatorship and establishing a democracy. However, the younger generations might feel more need to focus on the quality of the democracy they are living in. As such they might be harsher in condemning politicians’ wrongdoings, considering these practices part of a political culture that belongs to the past. According to this argument, corruption should decrease the probability of voting for a politician to a greater extent among younger citizens than among older citizens.
However, according to the life cycle effect argument, we could perhaps expect the opposite to be true. According to this argument, voters are likely to change the issues that they value most as they grow older. As the scarcity hypothesis tells us: “unmet physiological needs take priority over social, intellectual or esthetic needs” (Inglehart 1981:881)25. Young voters may be concerned with covering their material needs - such as getting a job, making ends meet and finding a home that they can afford - and may therefore be readier to trade off integrity against other valued aspects. Older individuals, having already met basic material objectives, may shift towards more post-material interests such as the
25 For this argument I only use the scarcity hypothesis. However, Inglehart’s theory of the shift towards post-material values is not based solely on this hypothesis. According to his theory, the societal shift towards post-material values is a combination of both the scarcity hypothesis and a socialization process that takes years to develop. It is therefore a combination of both cohorts and ageing effects.
118 Chapter 5
quality and integrity of those that govern them. According to this argument, we could expect voters in the oldest age group to be less willing to trade off corruption against other valued aspects.
As for as the moderating effect of partisanship, younger citizens should be less affected by partisan bias for two reasons. First, partisanship is an identity that is strengthened over time. As such, younger citizens should have weaker partisan ties than older citizens and therefore be less affected by the moderating effect of co-partisanship. Second, as younger cohorts entered the political arena in times of partisan de-alignment, during the mass mobilization of the ‘Indignados’ protest movement and during the appearance of new parties in the political arena (Vidal, 2018), it is to be expected that younger individuals have weaker partisan ties.
Studies that focus on the relationship between voters’ ages and their tolerance toward corruption are very rare and provide contradictory evidence. While some studies find a correlation between age and a lower tendency to justify corrupt activities (Torgler and Valev 2006; Gatti, Paternostro, and Rigolini 2003; Riera et al. 2013), others did not find this relationship (Anduiza, Gallego, and Muñoz 2013). This chapter assesses whether there are differences in the relative importance of corruption on voting choice between different age groups to provide further evidence to the state of the art.
Table 5.1 summarizes the expectations of the theories discussed regarding gender, political sophistication and age.
119 Individual heterogeneity
Table 5.1. Summary of expectations according to each theory
Individuals expected to punish Variable Theory corruption more harshly
Emotional and psychological differences Women Gender Opportunities and constraints None Higher awareness High sophisticates Sophistication More motivated reasoning Low sophisticates Cohort effects Younger individuals Age Life cycle effect Older individuals
5.3 Empirical strategy
As discussed in Chapter 2, due to the abstract nature of standard survey questions, there can be problems measuring the relative importance that respondents place on corruption. In this chapter, I use a conjoint experiment, especially designed and implemented for this dissertation, to identify whether the relative importance of corruption differs according to the gender, political sophistication or age of respondents. The analyses in this chapter draw on the same experiment discussed in Chapter 3. Please refer to section 3.3 for a detailed explanation of the survey and the design of the experiment.
Apart from the causal inferences discussed in Chapter 3, conjoint experiments also allow us to measure differences in subgroup preferences. While AMCEs are adequate for estimating the causal effect of the different attributes, these measures face challenges in identifying subgroup differences. Leeper, Hobolt, and Tilley (2019) demonstrate that marginal means, by also revealing preferences in the baseline condition, are more accurate in establishing subgroup differences. Furthermore, the nested model comparison, which compares the fit of one model that accounts for group differences and one that does not, allows us to establish whether the observed differences are statistically significant (for a more detailed explanation of these techniques see Leeper, Hobolt, and Tilley 2019). As such, in this chapter I estimate marginal means to assess whether the relative
120 Chapter 5
importance of corruption varies across the subgroups26. On another note, the theoretical discussion also set out the expectations that the conditioning effect of co-partisanship on corruption might be different across the subgroups. To test these hypotheses, we need to interact two attributes with each other and with the subgroup variables. To do that, I estimate the average component interaction effects (ACIE) including a triple interaction for each model (Hainmueller et al., 2014).
The gender and age of the respondents were measured with questions before the experiment, and political sophistication was measured with a set of four question that aimed to measure general political awareness. These questions were asked after the experiment to avoid influencing respondents’ answers in the experimental tasks. Asking a set of four political knowledge questions before the experiment could have influenced respondents’ motivation unequally and reduced their willingness to participate in the experimental tasks; this would have biased the data. Instead, it was done after the experiment, since respondents’ performance in the political knowledge questions would not be improved or worsened by participation in the experiment. In the first question, I showed respondents a picture of Jean-Claude Juncker and asked them to identify the person in the picture. In the second question, respondents were asked who the current Minister of Employment and Social Security was. In the third question, I asked them who was up for election in the European elections. In each case, respondents received a list of answer options and had to select one. Finally, they were asked to state the current unemployment rate in Spain.
Those respondents that answered zero questions correctly were coded as low sophisticates, those that answered between one and two correctly were coded as average sophisticates and those that answered three to four correctly were identified as high sophisticates. The age groups were coded into three ranges, one with respondents between 18 and 35, another with respondents between 36 and 55 and a third with all those over 55. Table 5.2 shows the distribution of respondents across the different genders, levels of political sophistication and age groups.
26 To do so, I use the cregg R package developed by Leeper (2019).
121 Individual heterogeneity
As for the dependent variable, respondents were asked to choose which mayor, of the pair they were shown, they would prefer for their municipality if there were local elections. In this chapter, I selected this dependent variable instead of their expressed probability of voting for candidates because the dichotomous outcome variable is more appropriate for estimating the marginal means.
Table 5.2. Distribution of respondents across gender, political sophistication and age groups
Cumulative Frequencies Percentage Percentage
Men 6156 51.17 51.17 Women 5874 48.83 100 Low sophistication 2.038 16.94 16.94 Average sophistication 5.622 46.73 63.67 High sophistication 4.370 36.33 100 18 to 35 years old 4168 34.65 34.65 36 to 55 years old 5924 49.24 83.89 Over 55 years old 1938 16.11 100
5.4 Results
The next section compares preferences between different subgroups. The marginal means shown in the following results can be interpreted as the degree of favorability toward politicians that have a particular characteristic.
Figure 5.1 compares responses of men (1) and women (2). Against the expectation of the theory that highlights their emotional and psychological differences, corruption does not have a stronger negative effect among women than among men, as both genders react equally strongly to corruption. Women seem to react somewhat more favorably to honest politicians than men, but the
122 Chapter 5
nested model comparison (Table 5.3), shows that this difference is not statistically significant.
Figure 5.2 compares low political sophisticates (1), to average (2) and high (3) sophisticates. In this case, the results confirm the expectations of the higher awareness theory, as a judge’s accusations of corruption have a stronger negative effect among highly sophisticated individuals than among low sophisticates. Similarly, low sophisticates also show less favorability towards honest politicians than average sophisticates and high sophisticates do. In addition, the nested model comparison confirms that these differences are statistically significant.
Finally, figure 5.3 shows the favorability towards a politician’s attributes between different age groups (1=18 to 35 years old, 2= 36 to 55, and 3= over 55). Interestingly, individuals from the youngest age group react less severely to accusation of corruption made by other parties than respondents from the older age ranges. However, all three groups have similar reactions when the accusations of corruption are made by a judge. In addition, respondents over 55 express more favorable opinions regarding co-partisan politicians than respondents from the younger age ranges. According to the nested model comparison, these differences are statistically significant. Considering the discussions put forwards in chapter 3, these results show that older individuals have a higher probability of trading integrity for co-partisanship than younger individuals do. In the next section, I analyze whether the moderating effect of co-partisanship also varies across age groups.
To sum up, the results of this chapter do not confirm gender differences in the probability of voting for corrupt politicians. However, they do confirm significant differences among people with different levels of political sophistication and of different age groups. Politically sophisticated individuals reject politicians accused of corruption by a judge more strongly than less sophisticated individuals. Young individuals react less strongly to accusations of corruption made by other parties than older individuals.
123 Individual heterogeneity
Figure 5.1. Comparison of marginal means for men (1) and women (1)
Figure 5.2. Comparison of marginal means for low sophisticates (1), average (2) and high (3) sophisticates
124 Chapter 5
Figure 5.3. Comparison of marginal means for respondents who are 18 to 35 years old (1), 36 to 55 years old (2) and over 55 years old (3)
Table 5.3. Nested model comparison
Residual Residual difference deviance Difference Deviance F P-value Gender 1 13209 2885.2 2 13202 2883.2 7 1.9947 1.3048 0.2433 Sophistication 1 13209 2885.2 2 13202 2881.1 7 4.1328 2.7054 0.08417*** Age 1 13209 2885.2 2 13202 2880.3 7 4.87 3.1894 0.002244*** ***p < 0.01 **p < 0.05 *p < 0.1 Model 1: Restricted model Model 2: Model that accounts for group differences
125 Individual heterogeneity
As discussed in the theoretical section of this chapter, another set of expectations argues that the conditioning effect of co-partisanship on corruption might be different across the subgroups. To test whether this is the case, I estimate the average marginal component interaction effects (AMCIEs) by fitting a linear regression and clustering standard error by respondents to a set of models that include a triple interaction between co-partisanship, corruption and the different subgroup variables (gender, age and political sophistication).
As shown in Tables 5.4 and 5.5, the results of the models (including the triple interaction) do not support either of the expectations regarding the different conditioning effect of co-partisanship. The empirical evidence does not show a different moderating effect of co-partisanship for gender, levels of political sophistication or age ranges.
126 Chapter 5
Table 5.4. Model including triple interaction with gender
Model 1 Choice
Woman 0.0373*** (0.00808) Low qualities -0.0628*** (0.00808) Weak performance -0.125*** (0.00808) Same party 0.159*** (0.0261) Corrupt -0.290*** (0.0129) Same party#Corrupt 0.0393 (0.0321) Woman 0.0293* (0.0151) Same party#Woman -0.00994 (0.0376) Corrupt#Woman -0.0335* (0.0186) Same party#Corrupt#Woman 0.00785 (0.0468) Constant 0.733*** (0.0126)
Observations 13,486 R-squared 0.121 Standard errors in parentheses Standard errors are clustered by respondent. *** p<0.01, ** p<0.05, * p<0.1
127 Individual heterogeneity
Table 5.5. Models including triple interaction with political sophistication and age
Model 2 Model 3 Choice Choice
Woman 0.0358*** Woman 0.0375*** (0.00816) (0.00808) Low qualities -0.0638*** Low qualities -0.0628*** (0.00816) (0.00808) Weak performance -0.125*** Weak performance -0.125*** (0.00816) (0.00808) Same party 0.180*** Same party 0.112*** (0.0463) (0.0310) Corrupt -0.229*** Corrupt -0.290*** (0.0214) (0.0156) Same party#Corrupt -0.0371 Same party#Corrupt 0.0632 (0.0571) (0.0388) Average sophistication 0.0593*** Age: 36-55 0.0108 (0.0206) (0.0167) High sophistication 0.0561*** Age: 65 and over -0.0160 (0.0215) (0.0226) Same party#Average sophistication -0.0398 Same party#36-55 0.0450 (0.0538) (0.0412) Same party#High sophistication -0.0252 Same party#56 and over 0.134** (0.0562) (0.0571) Corrupt#Average sophistication -0.0926*** Corrupt#36-55 -0.0289 (0.0254) (0.0205) Corrupt#High sophistication -0.102*** Corrupt#56 and over -0.0111 (0.0267) (0.0280) Same party#Corrupt#Average sophistication 0.0930 Same party#Corrupt#36-55 -0.0240 (0.0666) (0.0513) Same party#Corrupt#High Same party#Corrupt#56 and sophistication 0.102 over -0.0583 (0.0693) (0.0708) Constant 0.703*** Constant 0.745*** (0.0186) (0.0145)
Observations 13,216 Observations 13,486 R-squared 0.122 R-squared 0.121 Standard errors in parentheses Standard errors in parentheses Standard errors are clustered by respondent. Standard errors are clustered by respondent. *** p<0.01 ** p<0.05 * p<0.1 *** p<0.01 ** p<0.05 * p<0.1
128 Chapter 5
5.5 Conclusions
By exploiting the analysis of a conjoint experiment, this chapter draws an even more precise picture of the relative importance of corruption across different individual characteristics. The results confirm some significant differences between differing levels of political sophistication and age groups.
Politically sophisticated individuals reject politicians who are accused of corruption by a judge more strongly than less sophisticated individuals do. The results are in line with the awareness theory which suggests that more sophisticated individuals have more abilities to recognize corrupt activities, are more conscious of their negative consequences and are able to identify those responsible. However, the results do not support the motivated reasoning argument, according to which co-partisanship should have a stronger conditioning effect among high sophisticated individuals.
Nevertheless, it is important to highlight that this chapter is unable to control for another possible explanation that could be driving these results. The cognitive complexity of the experiment might have biased the results in favor of more sophisticated individuals, who needed less effort to read the information and complete the tasks. Yet again, this could also be a reflection of how difficult it is to navigate political information in real life.
As far as age is concerned, younger individuals react less harshly to accusations of corruption made by other parties than older individuals. However, these results could be a reflection of younger respondents’ lower trust in the information provided by other political parties, rather than a weaker reaction to corruption. In fact, all three age groups have similar reactions when the source of the corruption accusation is a judge. In turn, older individuals are more favorable towards co-partisan politicians than middle-aged and younger individuals are. However, the moderating effect of co-partisanship does not vary across age. In general, the results cannot support either a cohort effect or a life cycle effect, as the differences across age groups are only meaningful when accusations of corruption are made by other political parties but not when the accusation comes from a judge.
129 Individual heterogeneity
As far as gender is concerned, the results do not confirm differences supporting corrupt politicians, as both women and men react equally strongly to corruption. Furthermore, the moderating effect of co-partisanship on corruption is the same for both genders. These results seem to favor the theory that rejects women as having higher ethical standards than men: if this were the case, we would see corruption having a higher impact on voting choice among women.
Overall, this chapter confirms the results obtained in previous chapters: on average, voters do care about corruption, but when taking multidimensional decisions, they are willing to trade off integrity. The results also confirm some significant differences between differing levels of political sophistication and age groups. Nevertheless, a voter’s individual characteristics seem to play a rather moderate role in shaping the relative importance given to corruption.
130
Chapter 6
What about the other one? Characteristics of the alternative candidates
6.1 Introduction
This dissertation studies under what circumstances citizens decide to vote for a corrupt politician. So far, it has assessed what characteristics of candidates (Chapter 3) and what individual characteristics of voters (Chapter 5) shape the relative importance of corruption. Another key factor that determines whether voters hold corrupt politicians accountable is the supply of alternatives. When a preferred political choice is involved in corruption, the question of whether there are reasonable alternatives available or not will influence voters’ reactions (Charron & Bågenholm, 2016). Although not much research on corruption accountability focuses on the alternatives available, some studies show that the effective choices available determine voters’ punishment of corrupt politicians (Agerberg, 2020; Schleiter & Voznaya, 2014). However, I argue that it is not only a matter of having alternative political options available: these must also be attractive to voters. This chapter aims to go a step further and assess the characteristics of the alternatives that increase (or reduce) the punishment of corrupt incumbents at the polls.
131 Characteristics of the alternative
As a minimum requirement, the alternatives should at least seem more honest than the corrupt politician. Clearly, if the alternative candidates are perceived as equally corrupt, citizens will have no real incentive to hold politicians accountable for their wrongdoings (Pavão, 2015). As this dissertation focuses on contexts with medium to low levels of corruption, the scope is restricted to elections with enough competitors to ensure that there is always a clean alternative available.
Nevertheless, not only the perceived honesty of the other options should matter; other factors, such as these alternatives’ partisan identities or their ideologies, are likely to shape how voters react to corruption. The results obtained in Chapter 3 confirm that besides integrity, partisanship is the characteristic of candidates that most determines voting choice, and that co-partisanship strongly moderates the negative effect of corruption. This chapter draws on these results and addresses the question of how pervasive partisanship is. When faced with corruption in their preferred party, how do voters react if a clean alternative is available? I argue that voters’ preferences from among the other parties are likely to determine their reactions to corruption in their preferred party. Citizens have degrees of preference for all the parties in the political spectrum; they may identify more with one but they also have feelings about the other political parties. If the alternative belongs to a party that they feel fairly close to, they are more likely to punish the candidate from their preferred party.
However, as both research on economic voting and the results of Chapter 3 show, voters are not only driven by proximity to the political parties: the economic performance of politicians is also likely to determine their voting choice (Lewis‐Beck & Stegmaier, 2007). Therefore, I argue that the perceived attractiveness of the alternative is determined in terms of both party preferences and economic performance.
Furthermore, this chapter accounts for the fact that when voters learn about wrongdoings, they can hold the incumbent accountable in different ways. Thus far in this dissertation, punishing corrupt politicians has been operationalized as a decrease in the likelihood of voting for a politician, without considering what the voter’s actual behavioral response is. However, when voters learn about wrongdoings and they want to punish the corrupt politician in question, they can either switch to the alternative candidate or abstain from voting (De Vries
132 Chapter 6
& Solaz, 2017). Actually, results from both field and survey experiments show that corruption decreases turnout (Carreras & Vera, 2018; Chong et al., 2013). On the other hand, a cross-sectional analysis in Europe shows that voters switch to another political option if they have a reasonable alternative available (Charron & Bågenholm, 2016). However, they stick with the corrupt candidate if the alternative is too far away in ideological terms. In this chapter, I assess (i) when faced with corruption in their preferred political option, whether voters switch to a clean alternative or restrain from voting altogether and (ii) how the characteristics of the clean alternative determine their decisions.
I argue that the attractiveness of the alternative will shape voters’ reactions to corruption in their preferred political party; the degree of attractiveness is composed by both party preferences and economic performance. If the alternative seems to be reasonable in terms of party preferences and has shown strong economic performance, voters are more likely to punish the candidate from their preferred party for corruption. Nevertheless, if the alternative is too far away in terms of party proximity and had returned weak economic performance, voters are either going to stick with the corrupt politician or to abstain from voting.
6.2 Theoretical framework
As shown in Chapter 3, partisanship strongly determines respondents’ voting choice, and co-partisanship moderates the negative effect of corruption. Taking into account that partisanship is so determining, one would expect voters to stick with their preferred party even when it has been involved in corruption. However, Charron and Bågenholm (2016) show that voters are more likely to switch to another party if they have ideologically close alternatives available: what the authors call “reasonable alternatives”. Their result leads me to question to what extent the attenuating effect that partisanship has on corruption holds when voters have ideologically close alternatives available.
If voters stick with the corrupt co-partisans just because of their shared policy preferences, punishment of corruption at the polls should increase with the
133 Characteristics of the alternative
availability of reasonable alternatives. However, according to the original definition by the Michigan school, there is much more to partisanship than only ideological proximity. Campbell (1976) defined partisanship as social identification, a feeling of belonging to a certain group. According to this definition, partisan bonds are strong and do not just disappear with the supply of an alternative. If partisanship is driven mainly by in-group loyalties rather than policy preferences, then voters should stick with their preferred party no matter what alternatives are available. Furthermore, a lab experiment shows that individuals can support a corrupt politician just because they share the same artificially induced identity (Solaz et al., 2018).
I argue that partisanship is a mix of both policy preferences and in-group social identity and loyalty. In settings with enough political competitors, citizens have degrees of preference for the parties in the political spectrum (van der Eijk et al., 2006). They may identify with one party, but they also differentiate between the alternatives, liking some alternative political parties more than others. Like Charron and Bågenholm (2016), I expect voters to punish their preferred party for corruption if they have an alternative available that they perceive as reasonable, and not to punish their preferred party if the alternative is perceived as unreasonable. However, the “reasonable” verdict regarding the alternative is not only determined by ideological proximity to a party but also by broader feelings of likeability. Voters have opinions about all the parties in the political spectrum and whether they have a favorable position towards another party will determine if they switch to this alternative or not.
Notwithstanding, I argue that not only party preferences define the attractiveness of the alternative option. Chapter 3 and a long history of research on economic voting also reveal that economic performance influences voters’ support (for a review see Lewis‐Beck and Stegmaier 2007). Beyond the likeability of the alternative politician’s party, I also expect strong economic performance to determine whether voters switch to the clean alternative.
However, when voters find out about acts of corruption, they have more options than only voting for the clean alternatives: they can also abstain from voting altogether. Information about corruption could lead to less voting for several reasons. Research shows that finding out about corruption decreases trust in political institutions, politicians and civil servants (Ares & Hernández, 2017;
134 Chapter 6
Mishler & Rose, 2001). Corruption also reduces the legitimacy of the political system and the government (Andersen & Tverdova, 2003), and increases dissatisfaction with the functioning of democracy (Villoria et al., 2013). If corruption leads voters to distrust political institutions and the political system, then we could also expect them to be more likely to abstain. Empirical studies so far have indeed tended to find that corruption decreases turnout (Carreras & Vera, 2018; Chong et al., 2014; Sundström & Stockemer, 2015). I argue that the attractiveness of alternative candidates, defined by party preferences and economic performance, influences the probability of voters abstaining. Although corruption may have a strong effect on their trust in political institutions and politicians, this will be mitigated if there are attractive political options available. The more attractive the clean alternative is in terms of party preferences and economic performance, the less willing a voter will be to abstain from elections.
6.3 Empirical strategy
To test citizens’ reactions to clean alternatives, I use data from a conjoint experiment especially designed and implemented for this dissertation. By randomly assigning wrongdoings to different political parties, the experimental design excludes other possible confounding factors that may influence whether there are clean alternatives available and whether citizens choose to vote for them. The experiment was embedded in an online survey (n = 2275) in Spain in June 2016. The analyses in this chapter draw on the same experiment discussed in Chapter 3. Please refer to section 3.3 for a detailed explanation of the survey and the design of the experiment.
To measure respondents’ party preferences, I use respondents’ propensity to vote (PTVs) for each of the parties and combine it with the party label they see in the experimental profile. PTVs have been successfully used to measure citizens’ party preferences (Paparo et al., 2020; van der Eijk et al., 2006). PTVs were ascertained before the experiment by posing a question that asked respondents their propensity to vote for each of the parties on a scale from zero (“would never vote for her”) to 10 (“would definitely vote for her”) if there were
135 Characteristics of the alternative
municipal elections. This variable is recoded into three groups that cover the following categories: having a 70 to 100% propensity of voting for a party (baseline condition), a propensity from 40 to 60%, and one from 0 to 30%. The parties that were included in the experiment were the four most voted parties at the time the experiment was conducted: the PP (conservatives, in government when the experiment was conducted), the PSOE (social-democrats), Podemos (left) and Ciudadanos (center-right liberals) 27.
I use two strategies to assess how respondents react when their preferred party (the one they have a high propensity of voting for) is accused of corruption, and in addition, there is a clean alternative available. First, I fit a linear regression with clustered standard errors for respondents, with the probability of voting for each of the candidates seen in the experiment as a dependent variable. Respondents expressed their probability of voting for each candidate on a scale from zero (“would never vote for her”) to 10 (“would definitely vote for her”). The answers were rescaled from zero to one. I estimate the predicted probabilities of voting for different profiles of candidates and present a decision tree. Second, I create a new dependent variable that measures respondents’ reactions when confronted with a pair of candidates where one is corrupt and the other one is honest. Respondents can switch their vote to the clean alternative, abstain or stick with the corrupt candidate. This variable is coded by comparing the probability of voting for each of the two profiles seen in each of the rounds of the experiment. The decision is coded as vote for corrupt if the probability of voting is higher for the corrupt profile, as vote for clean if the likelihood of voting is higher for the clean alternative, and as abstain in those cases where the reported likelihood of voting for both profiles is zero. I run a multinomial logistic regression to assess which characteristics of the candidates determine whether voters switch to an alternative candidate, abstain or stick with the corrupt candidate.
27 Please refer to Chapter 3 for more details regarding the design of the experiment.
136 Chapter 6
6.4 Results
The decision tree in Figure 6.1 shows the predicted probabilities of voting for different profiles of candidates. The branches on the left show the predicted probabilities for a candidate of a party that respondents have a high propensity to vote for, while the other branches show the probabilities of voting for the clean alternative options. In the middle, there is a candidate belonging to a party that respondents have an average propensity of voting for; on the right, one belonging to a party that respondents have a low propensity of voting for.
Clearly, respondents have the highest probabilities of voting for a clean politician from their preferred party. However, when this politician is accused of corruption, they have higher probabilities of switching to a clean and reasonable alternative (a candidate that belongs to a party they have a propensity from 40 to 60 percent of voting for). Switching to the clean alternative is not as straightforward when the candidate belongs to a political party that they have a low propensity of voting for (from 0 to 30 percent). In this case, the probability of voting for a corrupt politician from the preferred party and a clean candidate proposed by an unreasonable political option is not significantly different.
Figure 6.1. Decision tree, predicted probabilities of voting for a candidate
Note: PTV refers to a respondent’s propensity to vote for a political party.
137 Characteristics of the alternative
As far as abstention is concerned, Table 6.1 shows the descriptive results for a variable that provides the reaction of respondents when confronted with a pair of candidates where one is corrupt and the other is honest. As we can see, in the majority of cases, respondents express a higher probability of voting for the clean alternative than voting for the corrupt one, or of abstaining. Nevertheless, in up to 30 percent of cases, respondents do prefer to abstain, while in 16 percent of cases respondents would rather vote for the corrupt candidate than for the clean alternative.
To assess under what conditions respondents would refrain from choosing the clean alternative, Table 6.2 shows the results of a multinomial logistic regression where the base outcome is voting for a clean candidate. Clearly, party preferences determine whether respondents decide to abstain, or to support the corrupt candidate. Respondents are more likely to abstain than voting for the clean alternative when they do not feel close to either the party of the clean candidate or to that of the corrupt one. Nevertheless, when the proximity to the party of the corrupt candidate increases, voting for this candidate is significantly more likely than voting for the clean alternative. If the clean alternative has a strong economic performance, this significantly decreases the likelihood of voting for the corrupt politician, while the alternative’s high qualities (university education and prolific management experience) decrease the voter’s likelihood of abstaining.
Table 6.1. Descriptive of Dependent Variable: choice
Percent N Vote clean 54.29 2920 Abstain 29.67 1596 Vote corrupt 16.04 863
138 Chapter 6
Table 6.2. Multinomial logit for choice when clean alternative is available
Abstention Vote corrupt PTV corrupt -0.10*** (0.01) 0.11*** (0.01) PTV clean -0.13*** (0.01) -0.11*** (0.02) Strong economic performance clean -0.10 (0.09) -0.31** (0.11) Strong economic performance corrupt -0.06 (0.09) 0.18 (0.11) High quality clean -0.21* (0.09) -0.16 (0.11) High quality corrupt -0.03 (0.09) 0.09 (0.11) Constant 0.18 (0.11) -1.16*** (0.14) Chi2 317.45 p 0.00 Observations 2863 Standard errors in parentheses * p<0.05, ** p<0.01, *** p<0.001 Note: Switching, abstaining or voting for corrupt candidate. Base category: Switching
6.5 Conclusions
The analysis of a conjoint experiment shows that the attractiveness of an alternative candidate explains voters’ reactions to corruption in their preferred political party. Respondents punish the corrupt politician from their preferred party by switching to an alternative, if this alternative belongs to a party they feel fairly close to or is known for her strong economic performance. However, respondents stick with the corrupt politician from their preferred party when the alternative belongs to a party they have a poor opinion of (a low propensity of voting for). Furthermore, respondents abstain from voting when they do not feel close to either the party of the corrupt politician or to the party of the clean alternative. All in all, the results of this chapter show that the way voters react to a corrupt politician from their preferred party depends on the characteristics of the alternative. Voters have clear preferences among the alternative options, and these determine whether they punish the corrupt politician or not.
These findings have clear implications for electoral accountability theories, as they show that accountability is more likely to work when numerous alternatives are available. Multi-party systems with many effective choices provide a wider variety of options, enabling voters to find an attractive preference to switch to. Conversely, in systems with only a few competitors, it is more difficult for voters
139 Characteristics of the alternative
to find an attractive alternative. Therefore, in party systems with not many effective choices, voters are more likely to abstain from voting or to stick to the corrupt politician when their preferred option is accused of corruption. As such, bipartisan systems undermine electoral accountability. Systems with only two competitors should therefore make sure that they implement other accountability mechanisms that are specifically designed to control corruption.
140
Chapter 7
Conclusions
7.1 Overview of main findings
This dissertation has analyzed the relative importance of corruption on voting choice, in order to understand under what circumstances corrupt politicians get reelected. This is a relevant topic because there are many examples all over the world of highly corrupt politicians being reelected. According to standard democratic theory, we should expect informed voters to punish corrupt politicians, as corruption is a clear sign of an underperforming government that will not act in the voters’ best interests (Fearon, 1999). However, empirical evidence shows that corruption is punished in elections only rather leniently (e.g. Dimock and Jacobson 1995; Chang, Golden, and Hill 2010; Peters and Welch 1978).
The all too frequent reelection of corrupt politicians has extended the idea that voters do not care much about a politician’s integrity (Fisman & Golden, 2017). However, the evidence derived from this dissertation rejects the idea of indifferent voters that do not care about corruption. If corrupt politicians survive elections, it is not because voters do not care about corruption, but because voters have multiple other concerns. Chapters 2, 3 and 4 provide strong
141 Conclusions
evidence that voters reject corruption and on average have low intentions of voting for a corrupt politician. Each of the three chapters used different empirical strategies to tackle this issue.
Chapter 1 used a list experiment to measure whether voters fail to consider corruption a motive for punishing a candidate from their preferred political party at the polls (either by abstaining or by voting for an alternative). Even when asked in an unobtrusive way, and therefore reducing the incentives to give socially desirable answers, a clear majority of respondents consider corruption a valid motive for punishing the politician. Chapter 2 used a conjoint experiment to measure the relative importance of corruption on voting choice, also considering other relevant factors, and shows that corruption has a strong negative effect on the reported probability of supporting a politician at the polls. Finally, Chapter 3 presented respondents with a dilemma in which bypassing the law leads to an optimal societal outcome. Even when faced with this dilemma, respondents have a higher propensity of voting for the mayor that follows the law and obtains a suboptimal outcome. Hence, the evidence of these chapters clearly shows that voters treasure politicians’ integrity and would ideally not vote for a corrupt politician.
Nevertheless, this dissertation also shows that there is a discrepancy between what voters would ideally like to do - punish corrupt politicians - and what they actually end up doing in elections. Although voters treasure integrity, this is not the only aspect they have to consider when casting a ballot, as they also value other characteristics of the candidates. When faced with multidimensional decisions, such as voting, voters may trade off a politician’s integrity against other characteristics that they also find desirable.
Chapter 1 shows that the expressed intention of voting for a corrupt candidate from the preferred party is substantially higher when the question is formulated as a tradeoff –referring to the politician’s management experience that residents value positively – in comparison to a simple question that does not include this tradeoff. However, the intention of voting does not increase when the same question is specifically formulated to increase respondents’ willingness to express their truthful position. Therefore, the main problem of standard survey questions that ask about attitudes towards corruption is not social desirability bias, but rather their inability to measure the relative importance of corruption
142 Chapter 7
in a setting of multidimensional decision making. In addition, Chapter 2 provides evidence that, when faced with a multidimensional decision, voters are indeed willing to trade off corruption for other valued characteristics such as partisan identity or performance. The experiment implemented in Chapter 2 shows that partisanship determines the vote to the same extent as corruption. Moreover, both co-partisanship and, to some extent, strong economic performance, moderate the negative effect corruption has on the vote.
The evidence compiled in this study shows that, although voters value politicians’ integrity and they would like to elect clean governments, in elections they may trade off integrity for other characteristics they value as well. In the experiments, I identified at least two tradeoffs voters engage with in elections: a tradeoff between integrity and partisanship and between integrity and a strong economic performance. It is intuitive to expect that voters face even more tradeoffs when it comes to real elections and that these tradeoffs can vary from election to election. This study had to limits its scope to a few tradeoffs; it remains for future research to test other possible tradeoffs that citizens face in elections.
The results clearly show that partisanship is the tradeoff that citizens engage with most strongly overall and is therefore likely to undermine holding politicians to account for corruption. According to my experiments, partisanship, together with corruption, is the attribute that most determines who a citizen votes for. Furthermore, co-partisanship strongly moderates the negative effect corruption has on the likelihood of voting for a politician.
It could be objected that the compelling results supporting the partisanship tradeoff were obtained in a very specific context, that it may not be possible to generalize these results to countries or regions where partisan ties are less strong. However, I would argue that they can indeed be extended to other contexts for various reasons. First, if anything, Spain could be considered a case with weak partisan ties, as Spanish citizens express very low trust in political parties. The manifestation of this low trust is not only expressed in surveys, but gave rise to the mass mobilizations and the ‘Indignados’ protest movement in 2011 (Anduiza et al., 2014; Vidal, 2018). Second, the data used in this dissertation was collected in a context in which new political parties had recently entered the political arena, and the results work just as well for these new parties as they do for the older,
143 Conclusions
more established parties. Third, as discussed in Chapter 6, the effects of partisanship in my data are a manifestation of both ideological preferences and group identity. Perhaps in some contexts, group attachments could be weaker than in Spain. However, voters will generally have certain ideological preferences that influence their voting decisions. Fourth, even if partisan attachments are weaker in some settings, research shows that there is no need for enduring or even pre-existing partisan ties for in-group loyalties to undermine corruption accountability. Even weak identities are effective in doing so. Participants in a lab experiment were willing to support a corrupt politician just because they shared the same identity as her, one that had been artificially induced at the beginning of the study (Solaz et al., 2018).
This dissertation has not only focused on what tradeoffs voters face when casting a vote, but it has also aimed to increase our understanding of how these tradeoffs work. This has been carried out by (i) identifying the causal mechanism that leads voters to vote (or not vote) for a malfeasant politician and (ii) exploring what voter’s characteristics and (iii) what characteristics of the alternative candidates increase the probability of voters trading off integrity for representation or competence.
Regarding the causal mechanism at play, I have shown that it is a drop in trust towards the corrupt politician that explains why voters may decide not to vote for her (Chapter 4). This reduction in trust is especially acute when voters get clear information about the private benefit the politician extracted from the corrupt activity. Therefore, if voters are not informed about this private benefit (and the wrongdoing can be justified as leading to superior societal outcomes), the politician’s trustworthiness does not decrease as much, and voters might keep voting for the dishonest politician. This study identified this causal mechanism, but future research should test it in different settings and determine the thresholds of private benefit that trigger the fall in trustworthiness.
As a matter of fact, the main findings of this dissertation – that voters care about corruption but in multidimensional decisions they are willing to trade off integrity – are in fact pervasive for different types of citizens. However, I am able to identify some modest but potentially relevant heterogeneities in citizens’ responses to corruption. As far as individual voter differences are concerned, my results show significant differences in the relative importance given to
144 Chapter 7
corruption for different levels of political sophistication (Chapter 5). Politically sophisticated individuals reject politicians accused of corruption by a judge more strongly than less sophisticated individuals do. Although these differences are statistically significant, they are quite modest and do not seem to be too consequential in substantive terms.
Regarding the characteristics of the politicians that could be voted for as alternatives, the evidence shows that these have a clear impact on voters’ reactions to corruption (Chapter 6). Individuals are willing to punish a corrupt politician from their preferred party by switching to an alternative if this is an attractive option: a candidate belonging to a party they feel fairly close to and/or is known for their strong economic performance. However, they stick with the corrupt politician of their preferred party when the alternative is unreasonable (i.e. belonging to a party they have a low propensity to vote for). Individuals abstain from voting when they feel close to neither the party of the corrupt politician nor to the party of the clean alternative.
It is important to note at this point that the results are derived from survey experiments. Although survey experiments are more successful for coming closer to respondents’ truthful attitudes and behavior than standard survey questions, we should bear in mind that these are still hypothetical settings. I used different techniques to reduce the artificiality of the experiments. Furthermore, the external validity of the main findings is ensured by the consistency of the results obtained with experiments using different treatments, measures and samples. However, as far as the exact estimates are concerned, I would recommend interpreting these with caution. As done in this conclusion, I encourage readers to always focus on the general results obtained, as the exact estimates obtained in the different experimental conditions can be sensitive to decision made in the experimental design.
It could also be objected that the information provided in the experiments is too credible to replicate real world scenarios. The facts about wrongdoings at the moment of casting a vote are never as straightforward as in the experimental conditions28. However, I argue that that the high credibility of the information
28 Judicial resolutions usually take years to materialize, as such when casting a ballot voters are usually only exposed to certain accusation of corruption.
145 Conclusions
about corruption is one strong point of this dissertation, as it poses a solid test for the tradeoff hypothesis. This study shows that even when obtaining highly credible information and, therefore, being certain that a candidate is corrupt, respondents are willing to trade integrity for other valued aspects. Hence, if in real settings, citizens receive less and lower quality information than in the experimental conditions of this study, we can only expect them to be even more willing to trade off integrity in real elections.
All in all, this dissertation provides compelling evidence that voters value politicians’ integrity and they would like to elect clean governments. But in elections they may trade off integrity for other valued characteristics. Consequently, the results show, among other things, that some of the failures of corruption accountability are not due to voters’ ignorance or to their indifference to politicians’ wrongdoings, but because integrity is not the only aspect that matters in their decision making. In the following sections, I discuss the implications of these findings for electoral accountability theories and for anti- corruption campaigns.
7.2 Implications of main findings for accountability
While the results of this dissertation do not entirely support a pessimistic interpretation, they also clearly point to the limits of elections as mechanisms for accountability. Although elections are a multidimensional phenomenon, when a government breaches principle of integrity, we expect voters to hold it accountable. But what if the incumbent has performed well on other issues? Should voters not reward them for that? As discussed by Manin, Przeworski, & Stokes (2012:50) “Governments make thousands of decisions that affect individual welfare; citizens have only one instrument to control these decisions: the vote. One cannot control a thousand targets with one instrument”. Cleary, an election every four or five years is not a strong enough instrument to hold politicians to account.
As shown in this dissertation, voters do worry about corruption and would ideally like to punish corrupt politicians. However, when casting a vote, they
146 Chapter 7
have many more aspects to consider. On one hand, voters may rationally choose to vote for a dishonest politician when this is the option that scores best in all the other dimensions that they are basing their decision on. On the other hand, voters are also unconsciously driven by psychological biases, such as partisan attachment, that moderate how they weigh up the alleged corruption perpetrated by a candidate.
Although this discussion points out the limits of elections as an accountability mechanism, the results do not completely disqualify this function of elections. The evidence shows that voters would prefer not to support corrupt politicians, but they might do so when other important issues are at stake. Therefore, the tradeoffs may be determined by the importance that individuals attribute to both corruption and to the other aspects considered when casting a vote. This leads to some space for higher electoral accountability, as certain circumstances can increase the saliency of corruption, and therefore shift the balance of the tradeoffs that voters face in elections.
However, since aggregate levels of corruption are correlated with poverty and economic inequality (Lambsdorff 2005), it is intuitive to think that the tradeoffs that voters face may be especially pronounced in those countries in which corruption is more widespread and accountability most needed. In poorer and more unequal countries, voters might be especially concerned with protecting their jobs, with access to quality education for their children or access to an effective health system, and they may therefore be more prone to trade off a politician’s integrity in order to ensure these other needs. As such, in those places where holding politicians accountable is most needed, voters may face even more pronounced tradeoffs that lead them to overlook corruption.
In addition, in contexts of widespread corruption, voters might face another obstacle to accountability; they may perceive all politicians as equally dishonest, either because all of them have been involved in malfeasance or because corruption is so prevalent that voters adopt cynical attitudes towards politics (Pavão, 2015). This clearly leaves them in a challenging position for holding corrupt politicians accountable.
An additional obstacle to corrupt politicians being held accountable at the polls, which has not been addressed in this dissertation but has been discussed by other
147 Conclusions
scholars, is the need for voters to coordinate in order to throw the malfeasants out of office. Even if a voter decides to punish the government for corruption, one vote by itself is not going to produce a different electoral outcome. In order to hold governments accountable, many voters have to decide to punish the incumbent for their wrongdoings. As such, individuals have to coordinate their vote around the clean alternative in order to hold the corrupt politician accountable (Fisman & Golden, 2017; Klašnja et al., 2018). Coordination refers to a “particular action an individual decides to make because of her knowledge or beliefs about the actions that other will choose” (Fisman and Golden 2017:2010). Therefore, an additional hurdle to accountability is an individual’s belief of what others will do. A voter that would like to punish a corrupt politician might not do so because of her expectations about the behavior of others.
Although, these conclusions may offer a gloomy view of elections, we should not go too far in disqualifying their usefulness for society. Elections might not work ideally as an accountability mechanism, but they are still the best instrument for making collective decisions (Przeworski, 2018). Assuming that we have to be governed, that no matter what, someone has to decide the rules that shape our everyday lives, elections are the best system for choosing those who govern. First, as it is well known, elections with universal secret suffrage allow everybody to express their will. An individual’s vote does not determine the result, but the joint sum of our wills does.
Second, elections create better incentives for those who govern than other forms of authorization to rule. For example, if rulers were selected through inheritance or lotteries they would “have no electoral incentives to behave well while in office” (Przeworski 2018: 115), because their chances of staying in office would be unrelated to their performance. As discussed in this dissertation, the positive incentives that elections produce are clearly also limited, especially when the payoffs from corruption are higher than the payoffs obtained from gaining reelection (Ashworth, 2012). Nevertheless, this does not mean that those in power would behave the same if they faced no threat whatsoever of being thrown out of office.
Third, voting reduces social conflict because losers accept the results of elections. Election results are considered legitimate because all people have had
148 Chapter 7
a chance to participate in the decision making process (Achen & Bartels, 2019). Even if this might not be a reason for some to comply, the fact that elections are a clear image of the distribution of preferences in society gives them a reason to do so. Losers are more willing to accept the results of elections because they realize they are a minority in society (Przeworski, 2018).
To sum up, this section has pointed out the limits of elections as an accountability mechanism. However, I do not aim to reject the utility of elections outright. Rather, my argument points out our own confused understanding of elections as an accountability instrument. As discussed, elections only work as an accountability mechanism in very specific situations, especially when corruption is an exceptionally salient issue, but might fail when accountability is needed the most. Elections may be the best instrument to select those who govern. But if our goal is to successfully control government, we should look for other mechanisms that have been conceived precisely for this purpose, such as institutional checks and balances. As such, the implications of this discussion are not that elections work unsatisfactorily, but rather that there is a problem regarding our own expectations of the purpose of elections.
7.3 Implications for anti-corruption campaigns
Most anti-corruption campaigns implemented by both international and national organizations promote transparency and information in order to increase citizens’ awareness (Pavão, 2015). For example, Transparency International puts a lot of effort into promoting individual agency in curbing corruption (Transparency International, n.d.). Numerous national governments have implemented transparency programs so that civil society can monitor politicians’ behavior. While information is essential for accountability, the results of this dissertation show that corruption accountability may be rather more complex.
The practical implications of the findings of this dissertation for the design of effective anti-corruption campaigns are twofold. On the one hand, programs engaged in increasing citizens’ awareness of corruption could serve to increase its saliency and therefore decrease the probability of voters trading it off. These
149 Conclusions
campaigns should, however, be specifically designed to help overcome tradeoffs. It is not enough to only denounce corruption; voters also need better alternatives to hold corrupt politicians to account. Nevertheless, this can only work successfully in countries with low levels of corruption. In countries with widespread corruption, finding out about politicians’ wrongdoings might actually increase citizens’ cynical attitudes towards politics. This is likely to lead them to overlook corruption when casting their votes (Pavão, 2015) − if all politicians are perceived as corrupt there is no need to vote on that issue − or to increase their probability of abstaining from elections and drive them out of institutional politics altogether.
If we are aware of the limits of electoral accountability and we really want to curb corruption, we should aim to strengthen horizontal accountability mechanisms, such as independent anti-corruption agencies. Following successful results in Hong Kong and Singapore, many countries have established anti-corruption agencies. However, the results are not always outstanding. The research has pointed out some successful strategies that anti-corruption agencies should adopt. They should (i) enforce “strong internal controls and accountability mechanisms”, (ii) “build alliances with citizens, state institutions, media, civil society, and international actors”, and (iii) implement “preventive efforts that disrupt corruption networks, together with educational efforts that reshape public norms and expectations” (Kuris 2014:3).
In sum, rather than strive to increase accountability for corruption at the polls, I consider that we should direct practitioners’ efforts to implementing successful institutional checks and balances. This does not mean that we should ignore transparency and the effective dissemination of information. These should be promoted as rights for citizens in democratic countries. Nevertheless, we should relieve citizens of the burden of controlling corruption in elections by ensuring it through other mechanisms. Indeed, if we already had successful horizontal accountability mechanisms in place, we would not have the problem of the failure of electoral accountability, because corrupt politicians would not be allowed to run for office in the first place (Pavão, 2018).
150 Chapter 7
7.4 Concluding remarks
A recurring account for why voters do not sanction corruption more harshly is that they are either ignorant − not properly informed about wrongdoings or unable to attribute responsibility − or that they are indifferent to corruption. This dissertation provides compelling evidence that voters value politicians’ integrity and they would like to elect clean governments. Nevertheless, holding politicians to account is not as simple as it looks. Besides integrity, voters consider many other important aspects when casting a ballot. Voting is a multidimensional decision and citizens may therefore trade off integrity for other valued aspects.
The results of this dissertation show that some of the failures of corruption accountability are not due to voters’ ignorance or indifference towards politicians’ wrongdoings, but because voters are forced to bring together multiple considerations in one sole decision. While elected representatives have the power to decide on almost any aspect of our lives, citizens have only one chance every few years to express their will. While governments take thousands of decisions, voters have just one ballot to express their opinion about these decisions. Considering this scenario, if we expect voters to always punish corrupt politicians, we are somewhat blinded by our innocence.
The argument of this dissertation rescues voters from an over-pessimistic view that is present also in the broader literature on political behavior, that posits that voters are too ignorant for democracy (for a review see Achen and Bartels 2019:9-12). If our system is not working for the vast majority of people, the problem is not the people but the system. In this case the problem is not elections themselves, but our own misleading understanding of elections as an instrument. Elections may be the best instrument for selecting those who govern, but if our goal is to successfully keep governments from being dishonest, we should look for other mechanisms that have been conceived precisely for this purpose, such as institutional checks and balances. As such, the reelection of corrupt politicians, or the broader lack of electoral accountability, should not be read as voters behaving unsatisfactorily, nor as elections working poorly, but rather as a problem of our own expectations regarding the very purpose of elections.
151
Appendices
Appendix A: Supplementary materials for Chapter 2
Table A.1. Randomization check: multinomial logit model
Direct question Long list Woman -0.03 (0.15) -0.08 (0.15) Age -0.01 (0.00) -0.01 (0.00) Middle education 0.38 (0.42) -0.61 (0.34) High education 0.45 (0.42) -0.61 (0.34) Income -0.01 (0.02) 0.03 (0.02) Ideology 0.01 (0.03) 0.01 (0.03) Partisanship -0.00 (0.00) -0.00 (0.00) Constant 0.02 (0.52) 0.66 (0.46) chi2 15.49 p 0.35 Observations 1190 Standard errors in parentheses Dependent variable: treatment assignment Base category: short list * p<0.05, ** p<0.01, *** p<0.001
152 Appendix A
Table A.2. Estimated Piecewise Proportions on the Treatment and Baseline/Control Lists as in Glynn (2013)
0 1 2 3 4 5 Sum 1 Treatment 0.055 0.0625 0.2025 0.4325 0.1575 0.09 1 2 Treatment "at least" 1 0.945 0.8825 0.68 0.2475 0.09 3 Control 0.05 0.1775 0.555 0.1375 0.08 0 1 4 Control “at least” 1 0.95 0.7725 0.2175 0.08 0 2 – 4 Joint 0 -0.005 0.11 0.4625 0.1675 0.09 0.825 (2 - 4)/1 Conditional 0 -0.08 0.54321 1.069364 1.063492 n/a Note: rows 1 and 3 show the proportion of respondents considering each particular number of items as reasons not to vote for the politician in question on the treatment and control lists. Rows 2 and 4 show the proportion that consider at least a particular number of items on the treatment and control lists. Row 5 displays the estimated difference between rows 2 and 4, which is an estimate of the proportion that do consider corruption a reason not to vote the politician and the total number of treatment list items indicated by the column. Row 6 presents the ratio of rows 5 and 1, which is an estimate of the proportion that consider corruption as a reason not to vote among those that report the total number of treatment list items indicated by the column.
Table A.3. Test to detect design effects as in Blair and Imai (2012)
Estimated population proportions SE pi(Y_i(0) = 0, Z_i = 1) -0.0050 0.0158 pi(Y_i(0) = 1, Z_i = 1) 0.1100 0.0264 pi(Y_i(0) = 2, Z_i = 1) 0.4625 0.0311 pi(Y_i(0) = 3, Z_i = 1) 0.1675 0.0255 pi(Y_i(0) = 4, Z_i = 1) 0.0900 0.0143 pi(Y_i(0) = 0, Z_i = 0) 0.0550 0.0114 pi(Y_i(0) = 1, Z_i = 0) 0.0675 0.0194 pi(Y_i(0) = 2, Z_i = 0) 0.0925 0.0314 pi(Y_i(0) = 3, Z_i = 0) -0.0300 0.0299 pi(Y_i(0) = 4, Z_i = 0) -0.0100 0.0197 Sensitive Item 1 0.5039624 Note: Bonferroni-corrected p-value. As this value is above alpha, we fail to reject the null hypothesis of no design effect.
153 Supplementary material for Chapter 2
Table A.4. Controlled multinomial logistic regression for tradeoff argument
Vote corrupt Abstain Reference= Simple question 0.00 (.) 0.00 (.) Tradeoff question 0.61** (0.19) 0.31 (0.16) Woman -0.15 (0.19) 0.05 (0.16) 30-39 years old -0.35 (0.28) -0.16 (0.25) 40-49 years old -0.20 (0.26) 0.03 (0.23) 50-65 years old -0.64* (0.31) -0.78** (0.28) Ideology 0.13** (0.05) 0.03 (0.04) Income -0.00 (0.00) 0.00* (0.00) Middle education -0.68 (0.40) -0.96** (0.34) High education -1.03* (0.40) -1.83*** (0.35) Constant -0.50 (0.50) 0.63 (0.43) chi2 78.64 p 0.00 Observations 864 Standard errors in parentheses Note: Dependent Variable: Vote for corrupt, vote for alternative or abstain. Base category: Vote for alternative * p<0.05, ** p<0.01, *** p<0.001
154
Appendix B: Supplementary material for Chapter 3
Table B.1. Descriptive of Dependent Variable: Vote probability
Treatment Mean Std Dev Observations Honest 0.49 0.35 4189 Accused by parties 0.27 0.3 4032 Accused by judge 0.22 0.29 4063 Total sample 0.33 0.34 12284
Table B.2. Average marginal component effects
Vote proability Se Ci low Ci high
Woman 0.0240*** (0.00534) 0.0135 0.0345 (Same party) Different party -0.260*** (0.00924) -0.278 -0.242 (High qualities) Low qualities -0.0288*** (0.00553) -0.0397 -0.0180 (Strong economic performance) Weak economic performance -0.0749*** (0.00580) -0.0863 -0.0636 (Honest) Accused parties -0.222*** (0.00769) -0.237 -0.207 Accused judge -0.266*** (0.00793) -0.281 -0.250
Observations 12,284 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Standard errors are clustered by respondent
155 Appendix B
B.1 Robustness check of relative weight hypothesis
To corroborate the results obtained when assessing the relative weight hypothesis, I re-ran model 1 with the partisanship variable coded differently; in this case I differentiated between the respondents that declared that they did not feel close to any of the parties and those that feel close to a different party. Therefore, we have three groups; those seeing the profile of a co-partisan candidate, those seeing the profile of a candidates belonging to a different party and non-partisans seeing the profile of a partisan candidate. The results observed in model 1 are corroborated; partisanship has an equally strong effect on the support for a candidate as corruption.
Figure B.1. Average marginal component effects (No partisanship)
Woman
Different party
No partisanship
Low qualities
Weak performance
Accused parties
Accused judge
-.3 -.2 -.1 0 .1
156 Supplementary material for Chapter 3
Table B.3. Average marginal component effects (No partisanship)
Vote Standard Ci low Ci high probability error Woman 0.0243*** (0.00533) 0.0139 0.0348 (Same party) Different party -0.252*** (0.00939) -0.271 -0.234 No partisanship -0.284*** (0.0126) -0.308 -0.259 (High qualities) Low qualities -0.0288*** (0.00552) -0.0397 -0.0180 (Strong economic performance) Weak economic performance -0.0745*** (0.00580) -0.0859 -0.0631 (Honest) Accused parties -0.223*** (0.00769) -0.238 -0.208 Accused judge -0.266*** (0.00791) -0.282 -0.251
Observations 12,284 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Standard errors are clustered by respondent
As a further corroboration test, I ran model 1 with the party preference variable instead of partisanship. Party preference is measured by combining the respondent’s propensity to vote for the party that is being assessed in the experiment (this information was acquired at the beginning of the questionnaire before the experiment). The results obtained with model 1 are corroborated, and actually in this case the effect of seeing the profile of a candidate that belongs to a party that the respondents had a very low propensity to vote for decreases the chances of voting for them significantly more than the accusation of corruption by a judge.
157 Appendix B
Figure B.2. Average marginal component effects (Party preferences)
Woman
PTV 4-6%
PTV 0-3%
Low qualities
Weak performance
Accused parties
Accused judge
-.3 -.2 -.1 0 .1
Note: PTV refers to a respondent’s propensity to vote for a political party
Table B.4. Average marginal component effects with party preferences Vote Standard Ci low Ci High probability error
Woman 0.0196*** (0.00514) 0.00948 0.0296 (Party with high propensity to vote) Party with average propensity to vote -0.117*** (0.0104) -0.137 -0.0968 Party with low propensity to vote -0.307*** (0.00865) -0.324 -0.290 High qualities Low qualities -0.0283*** (0.00525) -0.0386 -0.0180 (Strong economic performance) Weak economic performance -0.0771*** (0.00552) -0.0879 -0.0663 (Honest) Accused by parties -0.221*** (0.00732) -0.235 -0.206 Accused by judge -0.267*** (0.00763) -0.282 -0.252
Observations 12,284 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Standard errors are clustered by respondent
158 Supplementary material for Chapter 3
B.2 Robustness check of the conditional punishment hypothesis
As a robustness check of the results obtained when assessing the conditional punishment hypothesis, I present the results of a model that includes both interactions simultaneously.
The predicted probabilities and the semi-elasticities confirm the strong differential impact of corruption for when the candidate belongs to the same or to a different party as the respondent. Nevertheless, the results of the model that includes both interactions does not corroborate the moderating effect of the economic performance, as in this case there is no significant differential impact of corruption.
Table B.5. Predicted probabilities and the relative reduction of corruption (Model with both interactions)
Same party Different Party Reduction Reduction Vote Vote probability of vote probability probability of vote probability (%) (%) Honest 0.746 100 0.445 100 Corrupt 0.448 39.955 0.211 52.651
Weak Strong economic performance economic performance Reduction Reduction Vote Vote probability of vote probability probability of vote probability (%) (%) Honest 0.542 100 0.440 100 Corrupt 0.278 48.699 0.216 50.904
159 Appendix B
Table B.6. Derivatives expressed as a semi- elasticity (Model with both interactions)
ey/dx Contrast Partisanship Different party -0.755 (Reference) Same party -0.511 0.245*** Economic
performance Weak performance -0.747 (Reference) Strong performance -0.689 0.058
160 Supplementary material for Chapter 3
B.3 Conditional punishment of gender, education and management experience
As far as the moderating effect of gender, education and management experience is concerned, the predicted probabilities and the derivatives expressed as a semi-elasticity do not support that these characteristics condition punishment for corruption.
Table B.7. Predicted probabilities and the relative reduction of corruption
Man Woman Reduction Reduction
Vote probability of vote probability Vote probability of vote probability (%) (%) Honest 0.479 100 0.503 100 Corrupt 0.235 50.899 0.259 48.584
High education Low education and management experience management experience Reduction Reduction Vote probability of vote probability Vote probability of vote probability (%) (%) Honest 0.509 100 0.473 100 Corrupt 0.260 48.921 0.234 50.522
Table B.8. Derivatives expressed as a semi- elasticity
ey/dx Contrast Sex Woman -0.699 (Reference) Man -0.755 0.056 Education and management
experience High qualities -0.759 (Reference) Low qualities -0.716 0.043
161 Appendix B
B.4 Conditional punishment disentangling corruption accusations
Comparing both types of accusations, we see that, as we would expect, the differential impact of an accusation of corruption concerning a candidate of the same party and one from a different party is much greater when candidates are accused by other parties than when accused by a judge. That is, the moderating effect of partisanship on the negative effect of corruption is stronger when the information is uncertain and the sources are other political parties.
Table B.9. Predicted probabilities and the relative reduction of corruption
Same party Different Party Confidence Reduction Confidence Reduction Vote Interval of vote Vote Interval of vote
Probability probability probability probabilit Lower Higher Lower Higher (%) y (%) Honest 0.746 0.724 0.768 100 0.445 0.431 0.459 100 Accused parties 0.495 0.466 0.525 33.60 0.228 0.217 0.239 48.73 Accused judge 0.403 0.374 0.433 45.93 0.193 0.182 0.204 56.59
162 Supplementary material for Chapter 3
B.5 Further robustness checks
B.5.1 Analysis with individual fixed effects
To assess the robustness of the results presented in the chapter, I run a linear regression with individual fixed effects. Both the results of the relative weight hypothesis and the conditional punishment hypothesis are corroborated by this model.
Figure B.3. Average marginal component effects (with fixed effects)
Woman
Different party
Low qualities
Weak performance
Accused parties
Accused judge
-.3 -.2 -.1 0 .1
163 Appendix B
Table B.10. Average marginal component effects (with fixed effects)
Vote probability Se Ci low Ci high
Woman 0.0239*** (0.00519) 0.0137 0.0341 (Same party) Different party -0.240*** (0.00769) -0.255 -0.225 (High qualities) - Low qualitites -0.0317*** (0.00519) 0.0418 -0.0215 (Strong economic performance) - Weak economic performance -0.0860*** (0.00520) 0.0962 -0.0758 (Honest) Accused parties -0.242*** (0.00637) -0.255 -0.230 Accused judge -0.289*** (0.00637) -0.302 -0.277
Observations 12,284 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Standard errors are clustered by respondent
164 Supplementary material for Chapter 3
Table B.11.Predicted probabilities and the relative reduction of corruption (fixed effects)
Same party Different Party Reduction Reduction Vote Vote probability of vote probability probability of vote probability (%) (%) Honest 0.743 100 0.462 100 Corrupt 0.423 43.141 0.207 55.301
Strong Weak economic performance economic performance
Reduction Reduction Vote Vote probability of vote probability probability of vote probability (%) (%) Honest 0.565 100 0.446 100 Corrupt 0.274 51.468 0.205 54.005
Table B.12. Derivatives expressed as a semi- elasticity (fixed effects)
ey/dx Contrast Partisanship Different party -0.569 (Reference) Same party -0.827 0.258*** Economic performance Weak performance -0.752 (Reference) Strong performance -0.828 0.076**
165 Appendix B
B.5.2 Analysis only using the data of the first task
To rule out possible carryover effects that could bias the estimates reported in the second and third task, I report the same analysis here, but only taking into account the data of the first task that the respondents saw. The results again corroborate the relative weight hypothesis, as partisanship has an equally strong effect on the support for a candidate as the accusation by other parties.
These results also confirm the moderating effect of partisanship, as corruption has a significantly stronger negative effect when the candidate belongs to a different party. Nevertheless, the slight moderating effect of strong economic performance is not confirmed in this case, as the differential impact is not significant. However, this result does not allow us to reject the moderating effect of economic performance, since here I only employ one third of the data and therefore, the results could also be driven by the lower statistical power.
Figure B.4. Average marginal component effects (first task)
Woman
Different party
Low qualities
Weak performance
Accused parties
Accused judge
-.3 -.2 -.1 0 .1
166 Supplementary material for Chapter 3
Table B.13. Average marginal component effects (first task)
Vote probability Se Ci low Ci high
Woman 0.0317*** (0.00917) 0.0137 0.0497 (Same party) Different party -0.275*** (0.0145) -0.303 -0.246 (High qualities) Low qualitites -0.0335*** (0.00951) -0.0521 -0.0148 (Strong economic performance) Weak economic performance -0.0990*** (0.00946) -0.118 -0.0805 (Honest) Accused parties -0.237*** (0.0123) -0.261 -0.213 Accused judge -0.293*** (0.0120) -0.316 -0.269
Observations 4,076 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Standard errors are clustered by respondent.
Table B.14. Predicted probabilities and the relative reduction of corruption
Same party Different Party Reduction Reduction Vote Vote probability of vote probability probability of vote probability (%) (%) Honest 0.765 100 0.475 100 Corrupt 0.480 37.240 0.213 55.116
Strong economic performance Weak economic performance Reduction Reduction Vote Vote probability of vote probability probability of vote probability (%) (%) Honest 0.593 100 0.441 100 Corrupt 0.288 51.489 0.216 50.978
167 Appendix B
Table B.15. Derivatives expressed as a semi- elasticity
ey/dx Contrast Partisanship Different party -0.470 (Reference) Same party -0.829 0.359*** Economic performance Weak performance -0.768 (Reference) Strong performance -0.757 0.012
B.5.3 Relative weight hypothesis with forced choice as dependent variable
Finally, I report the results that use the forced choice as the dependent variable. In this case, respondents are forced to choose one of the two candidates instead of reporting their probability of voting for each one. The results of the linear regression clustered by respondents show that, in this case, the effect of corruption is significantly higher than the effect of partisanship. The slight difference between the distribution of these relative weights and the ones obtained with the probability of voting for a candidate as a dependent variable shows how the preferences of respondents change when they are obliged to choose between two candidates. Nevertheless, I consider the results of the probability of voting for a candidate more meaningful, as it gives respondents greater freedom to decide whether they would vote for a certain candidate or not. Therefore, the probability option provides a more accurate image of the way voters take decisions in real elections.
Conversely, the results of the conditional punishment hypothesis remain robust even when using the forced choice as a dependent variable. Actually, in this case, the moderating effect of economic performance is stronger and significant at a 99% confidence interval.
168 Supplementary material for Chapter 3
Figure B.5. Average marginal component effects (DV= Forced choice)
Woman
Different party
Low qualities
Weak performance
Accused parties
Accused judge
-.4 -.3 -.2 -.1 0 .1
Table B.16. Average marginal component effect (DV= Forced choice)
Forced choice Se Ci low Ci high
Woman 0.0347*** (0.00846) 0.0181 0.0513 (Same party) Different party -0.201*** (0.0110) -0.223 -0.179 (High qualities) Low qualitites -0.0634*** (0.00827) -0.0796 -0.0472 (Strong economic performance) Weak economic performance -0.137*** (0.00866) -0.154 -0.120 (Honest) Accused parties -0.271*** (0.0106) -0.292 -0.251 Accused judge -0.381*** (0.0103) -0.401 -0.361
Observations 12,284 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Standard errors are clustered by respondent.
169 Appendix B
Table B.17. Predicted probabilities and relative reduction of corruption
Same party Different Party Reduction Reduction
Vote probability of vote probability Vote probability of vote probability (%) (%) Honest 0.860 100 0.689 100 Corrupt 0.571 33.639 0.356 48.350
Strong Weak economic performance economic performance Reduction Reduction Vote probability of vote probability Vote probability of vote probability (%) (%) Honest 0.782 100 0.648 100 Corrupt 0.458 41.353 0.319 50.792
Table B.18. Derivatives expressed as a semi- elasticity
ey/dx Contrast Partisanship Different party -0.679 (Reference) Same party -0.415 0.264*** Economic performance Weak performance -0.730 (Reference) Strong performance -0.543 0.187***
170 Supplementary material for Chapter 3
B.6 Difference with other studies using conjoint experiments
Currently, other researchers are using conjoint experiments to assess accountability for corrupt politicians. I am aware of a study conducted by Klašnja, Lupu, and Tucker (2017), that was presented at the 2017 Annual Meeting of the European Political Science Association, where I also presented an earlier version of Chapter 3. Nevertheless, there are significant differences between the Klašnja et al. study and this chapter. First, the data for this chapter were collected in Spain in June 2016, while the Klašnja et al. study uses data gathered in May 2017 from Argentina, Chile, and Uruguay, which have completely different political contexts to Spain. Furthermore, the Klašnja et al. study focuses on the mitigating effects of corruption characteristics (how widespread it is and whether it produces side benefits) rather than on the candidates’ characteristics. To the best of my knowledge, Mares and Visconti (2018) are also working on a conjoint experiment in Romania to assess the mitigating effects of corruption characteristics (rather than candidates’ characteristics).
The only published article that also assesses candidates’ characteristics is a study by Franchino and Zucchini (2014). However, their study also differs substantially from my chapter. First, the Franchino and Zucchini (2014) study does not engage with the corruption accountability literature, and focuses on the valences and policy literature. In the experiment, they include a set of valence attributes (education, income and integrity) and a group of ideological attributes (spending in taxes and rights of same-sex couples), and assess how these factors interact with each other. Similarly to this chapter, they also find that ideological attributes mitigate the effects of integrity. In this chapter, I use party labels instead of actual policy positions: this allows me to control for the parties that respondents may infer when they receive information only on policies. This is an advantage in terms of the internal validity of the findings, as by referring explicitly to the party affiliation of the candidate, I control what party respondents think for every profile they see. It is also an improvement in terms of external validity, as this experimental setting resembles the settings of real elections more closely. In elections, voters will always receive information about the partisan attachment of candidates. In addition, the Franchino and Zucchini (2014) study does not assess the tradeoff hypothesis between economic
171 Appendix B
performance and the candidates’ integrity, which is one of the main contributions of this dissertation.
172 Supplementary material for Chapter 3
B.7 Questionnaire
(First screen)
En las próximas pantallas vamos a describir tres pares de perfiles de alcaldes de diferentes municipios con diferentes características.
Para cada par de perfiles encontrarás unas preguntas.
(On the next screens, we will describe three pairs of profiles of mayors from different municipalities with different characteristics.
For each pair of profiles, you will be asked some questions.)
173 Appendix B
(Second screen)
Por favor, lee con atención las características de cada perfil de alcalde (partido político al que pertenece, sexo, cualidades, resultados de su anterior mandato y características de su mandato) para responder con precisión a las preguntas que te haremos después.
(Please read the characteristics of each mayor’s profile (political party to which he/she belongs, gender, educational and managerial qualities, results of his/her previous term and characteristics of his/her mandate) carefully in order to answer with precision the questions that we will ask you later.)
174 Supplementary material for Chapter 3
P1. Imagina que hay elecciones en tu municipio. ¿Qué alcalde o alcaldesa preferirías para tu municipio?
(Imagine that there are elections in your municipality. Which mayor would you prefer for your municipality?)
P2. ¿Cuál es la probabilidad de que votaras al/la alcalde/sa García? Marca tu respuesta en esta escala, en la que el 0 significa que seguro que no lo/la votarías y el 10 que seguro que lo/la votarías.
(What is the probability that you would vote for Mayor García?
Mark your answer on this scale, where 0 means that you definitely would not vote for him/her and 10 means that you definitely would vote for him/her.)
P3. ¿Cuál es la probabilidad de que votaras al/la alcalde/sa Martínez? Marca tu respuesta en esta escala, en la que el 0 significa que seguro que no lo/la votarías y el 10 que seguro que lo/la votarías.
(What is the probability that you would vote for Mayor Martínez? Mark your answer on this scale, where 0 means that you definitely would not vote for him/her and 10 that you definitely would vote for him/her.)
175
Appendix C: Supplementary material for Chapter 4
Table C.1. Original experiment wording
Common introduction to the vignette: Unas lluvias torrenciales han inundado y dañado gravemente el alcantarillado de una ciudad. Esta situación produce grandes inconvenientes para los residentes de la ciudad. Para realizar las reparaciones y resolver el problema el ayuntamiento tiene que seleccionar una empresa a través de un concurso público. Sin embargo, convocar un concurso retrasará la resolución del problema. La única forma de acelerar la reparación es evitar el concurso público y asignar el contrato directamente a una empresa con experiencia. Sin embargo, esto no sería legal.
Treatments: Legal: David G. P., alcalde de la ciudad y miembro de/l {respondent’s preferred party}, convoca un concurso público que finalmente gana una empresa con experiencia. Esta decisión respeta el procedimiento, aunque retrasa las reparaciones. Criminal: David G. P., alcalde de la ciudad y miembro de/l {respondent’s preferred party}, no convoca el concurso y asigna el contrato a una empresa con experiencia. Esta decisión acelera las reparaciones, aunque no respeta el procedimiento. Corrupt: David G. P., alcalde de la ciudad y miembro de/l {respondent’s preferred party}, no convoca el concurso y asigna el contrato a una empresa con experiencia que ha colaborado en la campaña electoral de su partido. Esta decisión acelera las reparaciones, aunque no respeta el procedimiento.
176 Appendix C
C.1 Manipulation checks
To check whether the treatment is successful in describing a gradual treatment of malfeasance, I asked respondents about the extent to which they think that the mayor they saw in the experiment is malfeasant (10 being completely corrupt). The results clearly show that the experiment is successful in tapping into degrees of malfeasance. The legal mayor is considered as the least malfeasant and this perception gradually increases until arriving at the corrupt mayor.
Figure C.1. Mean perceived malfeasance across the different treatment groups
10
9
8
7
6
5
4
Perceivedmalfeasance 3
2
1
0
Legal decision Criminal decision Corrupt decision
Note: 95% confidence intervals around the means
177 Supplementary material for Chapter 4
C.2 Additional mediation analyses
Figure C.2. Mediation analysis of empathy with confounding by an alternative mechanism (trustworthiness)
ACME
ADE
Total Effect Criminal vs Legal Corrupt vs Criminal Corrupt vs Legal
-4 -3 -2 -1 0 Effects on probability to vote Percent mediated by empathy: Legal vs Criminal -10.57% / Criminal vs Corrupt 25.24% / Legal vs Corrupt 9.91%
178 Appendix C
C.3 Mediation sensitivity analyses
Figure C.3. Sensitivity tests of multiple mediation analyses
Trustworthiness mediation Efficiency mediation
Criminal vs. Legal Criminal vs. Legal
Corrupt vs. Criminal Corrupt vs. Criminal
179 Supplementary material for Chapter 4
Corrupt vs. Legal Corrupt vs. Legal
180 Appendix C
C.4. Average treatment effects including non-partisan attachment
In this section, I replicate the average treatment effect including respondents that declare that they do not feel close to any of the parties. These respondents were randomly assigned to one of the four parties with more representation in the Spanish parliament (PP, PSOE, Podemos and Ciudadanos) at the time the experiment was conducted.
Figure C.4. Mean propensity to vote across different treatment conditions
10
9
8
7
6
5
4
Meanpropensity to vote 3
2
1
0
Legal decision Criminal decision Corrupt decision
Note: 95% confidence intervals around the means
181 Supplementary material for Chapter 4
Figure C.5. Traits analyses: Proportion of respondents considering the mayor trustworthy and efficient across different treatments conditions
1
.8
.6
Proportion .4
.2
0 Politician is trustworthy Politician is efficient
Legal decision Criminal decision Corrupt decision
Note: 95% confidence intervals around the means
182
Bibliography
Achen, C. H., & Bartels, L. M. (2019). Democracy for Realists. In Democracy for Realists. Princeton University Press. https://doi.org/10.2307/j.ctvc7770q
Agerberg, M. (2020). The Lesser Evil? Corruption Voting and the Importance of Clean Alternatives. Comparative Political Studies, 53(2), 253–287. https://doi.org/10.1177/0010414019852697
Allen, N., & Birch, S. (2011). Political Conduct and Misconduct: Probing Public Opinion. Parliamentary Affairs, 64(1), 61–81. https://doi.org/10.1093/pa/gsq007
Allen, N., Birch, S., & Sarmiento-Mirwaldt, K. (2016). Honesty above all else? Expectations and perceptions of political conduct in three established democracies. Comparative European Politics, 1–24. https://doi.org/10.1057/s41295-016-0084-4
Amundsen, I. (1999). Political corruption: An introduction to the issues. CMI Working Paper, WP 1999:7.
Andersen, C. J., & Tverdova, Y. V. (2003). Corruption, political allegiances, and attitudes toward government in contemporary democracies. American Journal of Political Science, 47(1), 91–109. https://doi.org/10.1111/1540-5907.00007
Anduiza, E., Cristancho, C., & Sabucedo, J. M. (2014). Mobilization through online social networks: the political protest of the indignados in Spain.
183 Bibliography
Information Communication and Society, 17(6), 750–764. https://doi.org/10.1080/1369118X.2013.808360
Anduiza, E., Gallego, A., & Muñoz, J. (2013). Turning a Blind Eye: Experimental Evidence of Partisan Bias in Attitudes Toward Corruption. Comparative Political Studies, 46(12), 1664–1692. https://doi.org/10.1177/0010414013489081
Ares, M., & Hernández, E. (2017). The corrosive effect of corruption on trust in politicians: Evidence from a natural experiment. Research & Politics, 4(2), 205316801771418. https://doi.org/10.1177/2053168017714185
Ashworth, S. (2012). Electoral Accountability: Recent Theoretical and Empirical Work. Annual Review of Political Science, 15(1), 183–201. https://doi.org/10.1146/annurev-polisci-031710-103823
Bågenholm, A. (2013). Throwing the rascals out? The electoral effects of corruption allegations and corruption scandals in Europe 1981–2011. Crime, Law and Social Change, 60(5), 595–609. https://doi.org/10.1007/s10611-013-9482-6
Barnes, T. D., & Beaulieu, E. (2014). Gender Stereotypes and Corruption: How Candidates Affect Perceptions of Election Fraud. Politics & Gender, 10(03), 365–391. https://doi.org/10.1017/S1743923X14000221
Beaulieu, E. (2014). From voter ID to party ID: How political parties affect perceptions of election fraud in the U.S. Electoral Studies, 35, 24–32. https://doi.org/10.1016/J.ELECTSTUD.2014.03.003
Birch, S., & Allen, N. (2015). Judging politicians: The role of political attentiveness in shaping how people evaluate the ethical behaviour of their leaders: Judging politicians. European Journal of Political Research, 54(1), 43–60. https://doi.org/10.1111/1475-6765.12066
Blair, G., & Imai, K. (2012). Statistical Analysis of List Experiments. Political Analysis, 20(1), 47–77. https://doi.org/10.1093/pan/mpr048
Boas, T. C., Hidalgo, F. D., & Melo, M. A. (2018). Norms versus Action: Why Voters Fail to Sanction Malfeasance in Brazil. American Journal of Political Science, ajps.12413. https://doi.org/10.1111/ajps.12413
184 Bibliography
Botero, S., Cornejo, R. C., Gamboa, L., Pavao, N., & Nickerson, D. W. (2015). Says Who? An Experiment on Allegations of Corruption and Credibility of Sources. Political Research Quarterly, 68(3), 493–504. https://doi.org/10.1177/1065912915591607
Braungart, R. G., & Braungart, M. M. (1986). Life-Course and Generational Politics. In Annual Review of Sociology (Vol. 12, pp. 205–231). Annual Reviews. https://doi.org/10.2307/2083201
Breitenstein, S. (2019). Choosing the crook: A conjoint experiment on voting for corrupt politicians. Research & Politics. https://doi.org/10.1177/2053168019832230
Broidy, L., Cauffman, E., Espelage, D. L., Mazerolle, P., & Piquero, A. (2003). Sex Differences in Empathy and Its Relation to Juvenile Offending. Violence and Victims, 18(5), 503–516. https://doi.org/10.1891/vivi.2003.18.5.503
Brollo, F. (2012). Why Do Voters Punish Corrupt Politicians? Evidence from the Brazilian Anti-Corruption Program. In SSRN Electronic Journal. https://doi.org/10.2139/ssrn.2141581
Bullock, J. G., Green, D. P., & Ha, S. E. (2010). Yes, But What’s the Mechanism? (Don’t Expect an Easy Answer). Journal of Personality and Social Psychology, 98(4), 550–558. https://doi.org/10.1037/a0018933
Campbell, A. (1976). The American voter. The University of Chicago Press.
Carreras, M., & Vera, S. (2018). Do Corrupt Politicians Mobilize or Demobilize Voters? A Vignette Experiment in Colombia. Latin American Politics and Society, 60(3), 77–95. https://doi.org/10.1017/lap.2018.25
Chang, E. C. C., Golden, M. A., & Hill, S. J. (2010). Legislative Malfeasance and Political Accountability. World Politics, 62(2), 177–220.
Charron, N., & Bågenholm, A. (2016). Ideology, party systems and corruption voting in European democracies. Electoral Studies, 41, 35–49. https://doi.org/10.1016/j.electstud.2015.11.022
Chong, A., Ana, L., Karlan, D., & Wantchekon, L. (2013). Looking Beyond the Incumbent: Exposing Corruption and the Effect on Electoral Outcomes. NBER Working Paper, 17679.
185 Bibliography
Chong, A., De La O, A. L., Karlan, D., & Wantchekon, L. (2014). Does Corruption Information Inspire the Fight or Quash the Hope? A Field Experiment in Mexico on Voter Turnout, Choice, and Party Identification. The Journal of Politics, 77(1). https://doi.org/10.1086/678766
Corstange, D. (2009). Sensitive Questions, Truthful Answers? Modeling the List Experiment with LISTIT. Political Analysis, 17(1), 45–63. https://doi.org/10.1093/pan/mpn013
Corstange, D. (2010). Vote buying under competition and monopsony: Evidence from a list experiment in lebanon. Presentation at the Annual Meeting of the American Political Science Association, Washington, DC.
Costas-Pérez, E., Solé-Ollé, A., & Sorribas-Navarro, P. (2009). Do voters really tolerate corruption ? Evidence from Spanish Mayors. 102, 1–43.
Costas-Pérez, E., Solé-Ollé, A., & Sorribas-Navarro, P. (2012). Corruption scandals, voter information, and accountability. European Journal of Political Economy, 28(4), 469–484. https://doi.org/10.1016/j.ejpoleco.2012.05.007
Cushman, F. (2013). Action, Outcome, and Value: A Dual-System Framework for Morality. Personality and Social Psychology Review, 17(3), 273–292. https://doi.org/10.1177/1088868313495594
Dahl, R. A. (1971). Polyarchy : participation and opposition. Yale University Press.
De Vries, C. E., & Solaz, H. (2017). The Electoral Consequences of Corruption. Annual Review of Political Science, 20(1), 391–408. https://doi.org/10.1146/annurev-polisci-052715-111917
Dimock, M. A., & Jacobson, G. C. (1995). Checks and Choices: The House Bank Scandal’s Impact on Voters in 1992. The Journal of Politics, 57(4), 1143–1159. https://doi.org/10.2307/2960406
Dollar, D., Fisman, R., & Gatti, R. (2001). Are Women Really the. Journal of Economic Behavior & Organization, 46(4), 423–429.
Dunning, T., Grossman, G., Humphreys, M., Hyde, S. D., McIntosh, C., Nellis, G., Adida, C. L., Arias, E., Bicalho, C., Boas, T. C., Buntaine, M. T., Chauchard, S., Chowdhury, A., Gottlieb, J., Daniel Hidalgo, F., Holmlund, M., Jablonski, R., Kramon, E., Larreguy, H., … Sircar, N.
186 Bibliography
(2019). Voter information campaigns and political accountability: Cumulative findings from a preregistered meta-analysis of coordinated trials. Science Advances, 5(7). https://doi.org/10.1126/sciadv.aaw2612
Eggers, A. C., & Fisher, A. (2011). Electoral Accountability and the UK Parliamentary Expenses Scandal : Did Voters Punish Corrupt MPs? https://doi.org/http://dx.doi.org/10.2139/ssrn.1931868
Eggers, A. C., Vivyan, N., & Wagner, M. (2018). Corruption, Accountability, and Gender: Do Female Politicians Face Higher Standards in Public Life? The Journal of Politics, 80(1), 321–326. https://doi.org/10.1086/694649
Esaiasson, P., Muñoz, J., & Fernández-Vázquez, P. (2014). Roba per hace? An experimental test of the competence-corruption tradeoff hy- pothesis in Spain and Sweden (QoG Working Paper Series 2014:02).
Esarey, J., & Chirillo, G. (2013). Fairer sex or purity myth? Corruption, gender, and institutional context. In Politics and Gender (Vol. 9, Issue 4, pp. 361–389). Cambridge University Press. https://doi.org/10.1017/S1743923X13000378
Esarey, J., & Schwindt-Bayer, L. A. (2018). Women’s Representation, Accountability and Corruption in Democracies. In British Journal of Political Science (Vol. 48, Issue 3, pp. 659–690). Cambridge University Press. https://doi.org/10.1017/S0007123416000478
Everett, J. A. C., Pizarro, D. A., & Crockett, M. J. (2016). Inference of trustworthiness from intuitive moral judgments. Journal of Experimental Psychology: General, 145(6), 772–787. https://doi.org/10.1037/xge0000165
Fearon, J. D. (1999). Electoral Accountability and the Control of Politicians: Selecting Good Types versus Sanctioning Poor Performance. In A. Przeworski, S. C. Stokes, & B. Manin (Eds.), Democracy, Accountability, and Representation (pp. 55–97). Cambridge University Press. https://doi.org/10.1017/CBO9781139175104.003
Ferejohn, J. (2012). Accountability and Authority: Toward a Theory of Political Accountability. In Democracy, Accountability, and Representation (pp. 131–153). Cambridge University Press. https://doi.org/10.1017/cbo9781139175104.005
187 Bibliography
Fernández-Vázquez, P., Barberá, P., & Rivero, G. (2016). Rooting Out Corruption or Rooting For Corruption ? The Heterogeneous Electoral Consequences of Scandals. Political Science Research and Methods, 4(2). https://doi.org/10.1017/psrm.2015.8
Ferraz, C., & Finan, F. (2008). Exposing Corrupt Politicians: The Effects of Brazil’s Publicly Released Audits on Electoral Outcomes. The Quarterly Journal of Economics, 123(2), 703–745. https://doi.org/10.1162/qjec.2008.123.2.703
Ferrer, S. (2020). Responsibility attribution for corruption scandals. Local Government Studies, 46(1), 22–43. https://doi.org/10.1080/03003930.2019.1583560
Fisman, R., & Golden, M. A. (2017). Corruption : what everyone needs to know. Oxford University Press.
Franchino, F., & Zucchini, F. (2014). Voting in a Multi-dimensional Space: A Conjoint Analysis Employing Valence and Ideology Attributes of Candidates. Political Science Research and Methods, 3(02), 221–241. https://doi.org/10.1017/psrm.2014.24
Funk, C. L. (1996). The Impact of Scandal on Candidate Evaluations: An Experimental Test of the Role of Candidate Traits. Political Behavior, 18(1), 1–24.
Garzia, D. (2011). The personalization of politics in Western democracies: Causes and consequences on leader–follower relationships. The Leadership Quarterly, 22(4), 697–709. https://doi.org/10.1016/j.leaqua.2011.05.010
Gatti, R., Paternostro, S., & Rigolini, J. (2003). Individual Attitudes toward Corruption: Do Social Effects Matter? (Policy Research Working Papers). The World Bank. https://doi.org/10.1596/1813-9450-3122
Gerber, A. S., & Green, D. P. (2012). Field experiments : design, analysis, and interpretation. W.W. Norton.
Giné, X., & Mansuri, G. (2018). Together We Will: Experimental Evidence on Female Voting Behavior in Pakistan. American Economic Journal: Applied Economics, 10(1), 207–235. https://doi.org/10.1257/app.20130480
188 Bibliography
Glynn, A. N. (2013). What Can We Learn with Statistical Truth Serum? Public Opinion Quarterly, 77(S1), 159–172. https://doi.org/10.1093/poq/nfs070
Golden, M. A. (2010). Some Puzzles of Political Corruption in Modern Advanced Democracies. In Democracy and Accountability: Globalized Political Responsibility. Toyko: Fukosha, 184–199.
Gonzalez-Ocantos, E., de Jonge, C. K., Meléndez, C., Osorio, J., & Nickerson, D. W. (2012). Vote Buying and Social Desirability Bias: Experimental Evidence from Nicaragua. American Journal of Political Science, 56(1), 202–217. https://doi.org/10.1111/j.1540- 5907.2011.00540.x
Green, D. P., Ha, S. E., & Bullock, J. G. (2010). Enough Already about “Black Box” Experiments: Studying Mediation Is More Difficult than Most Scholars Suppose. The ANNALS of the American Academy of Political and Social Science, 628(1), 200–208. https://doi.org/10.1177/0002716209351526
Gupta, S., Davoodi, H., & Alonso-Terme, R. (2002). Does corruption affect income inequality and poverty? Economics of Governance, 3(1), 23–45. https://doi.org/10.1007/s101010100039
Hainmueller, J., Hopkins, D. J., & Yamamoto, T. (2014). Causal inference in conjoint analysis: Understanding multidimensional choices via stated preference experiments. Political Analysis, 22(1), 1–30. https://doi.org/10.1093/pan/mpt024
Healy, A., & Malhotra, N. (2013). Retrospective Voting Reconsidered. Annual Review of Political Science, 16(1), 285–306. https://doi.org/10.1146/annurev-polisci-032211-212920
Holbrook, A. L., & Krosnick, J. A. (2010). Social desirability bias in voter turnout reports Tests using the item count technique. Public Opinion Quarterly, 74(1), 37–67. https://doi.org/10.1093/poq/nfp065
Imai, K., Keele, L., Tingley, D., & Yamamoto, T. (2011). Unpacking the Black Box of Causality: Learning about Causal Mechanisms from Experimental and Observational Studies. American Political Science Review, 105(04), 765–789. https://doi.org/10.1017/S0003055411000414
Imai, K., & Yamamoto, T. (2013). Identification and Sensitivity Analysis for
189 Bibliography
Multiple Causal Mechanisms: Revisiting Evidence from Framing Experiments. Political Analysis, 21(02), 141–171. https://doi.org/10.1093/pan/mps040
Incerti, T. (2019). Corruption information and vote share: A meta-analysis and lessons for survey experiments.
Inglehart, R. (1981). Post-Materialism in an Environment of Insecurity. American Political Science Review, 75(4), 880–900. https://doi.org/10.2307/1962290
Janus, A. L. (2010). The influence of social desirability pressures on expressed immigration attitudes. Social Science Quarterly, 91(4), 928–946.
Jareño Leal, Á. (2017). Conductas Delictivas en Materia de Contratación Pública.
Jiménez, F. (2009). Building Boom and Political Corruption in Spain. South European Society and Politics, 14(3), 255–272. https://doi.org/10.1080/13608740903356541
Kam, C. D. (2005). Who Toes the Party Line? Cues, Values, and Individual Differences. Political Behavior, 27(2), 163–182. https://doi.org/10.1007/s11109-005-1764-y
Klašnja, M., Little, A. T., & Tucker, J. A. (2018). Political Corruption Traps. Political Science Research and Methods, 6(3), 413–428. https://doi.org/10.1017/psrm.2016.45
Klašnja, M., Lupu, N., & Tucker, J. A. (2017). When Do Voters Sanction Corrupt Politicians?
Klašnja, M., & Tucker, J. A. (2013). The economy, corruption, and the vote: Evidence from experiments in Sweden and Moldova. Electoral Studies, 32(3). https://doi.org/10.1016/j.electstud.2013.05.007
Konstantinidis, I., & Xezonakis, G. (2013). Sources of tolerance towards corrupted politicians in Greece: the role of trade offs and individual benefits. Crime, Law and Social Change, 60(5), 549–563. https://doi.org/10.1007/s10611-013-9478-2
Kpundeh, S. J. (1998). Political will in fighting corruption. Corruption and Integrity Improvement Initiatives in Developing Countries, 91–110.
190 Bibliography
Kramon, E., & Weghorst, K. (2019). (Mis)Measuring Sensitive Attitudes with the List Experiment. Public Opinion Quarterly, 83(S1), 236–263. https://doi.org/10.1093/poq/nfz009
Kreps, T. A., & Monin, B. (2014). Core Values Versus Common Sense Consequentialist Views Appear Less Rooted in Morality. Personality and Social Psychology Bulletin, 40(11), 1529–1542. https://doi.org/10.1177/0146167214551154
Krosnick, J. A. (1988). The Role of Attitude Importance in Social Evaluation: A Study of Policy Preferences, Presidential Candidate Evaluations, and Voting Behavior. Journal of Personality and Social Psychology, 55(2), 196–210. https://doi.org/10.1037/0022-3514.55.2.196
Kruttschnitt, C. (1994). Gender and interpersonal violence. In J. Roth & A. Reiss (Eds.), Understanding and preventing violence: Social influences (Vol.3, pp. 295–378). National Academy of Sciences.
Kuklinski, J. H., Cobb, M. D., & Gilens, M. (1997). Racial Attitudes and the “New South.” The Journal of Politics, 59(02), 323–349. https://doi.org/10.2307/2998167
Kurer, O. (2001). Why do voters support corrupt politicians? In A. K. Jain (Ed.), The political economy of corruption (pp. 63–86). Routledge.
Kuris, G. (2014). From Underdogs to Watchdogs: How Anti-corruption Agencies Can Hold Off Potent Adversaries. In From Underdogs to Watchdogs: How Anti-Corruption Agencies Can Hold Off Potent Adversaries.
Lambsdorff, G. (2005). Consequences and causes of corruption: What do we know from a cross-section of countries? Standard- Nutzungsbedingungen. Passauer Diskussionspapiere, Volkswirtschaftliche Reihe, 35.
Laustsen, L., & Bor, A. (2017). The relative weight of character traits in political candidate evaluations: Warmth is more important than competence, leadership and integrity. Electoral Studies, 49, 96–107. https://doi.org/10.1016/j.electstud.2017.08.001
Leeper, T. (2019). Cregg: Simple Conjoint Analyses and Visualization. R package version 0.3.1.
Leeper, T. J., Hobolt, S. B., & Tilley, J. (2019). Measuring Subgroup
191 Bibliography
Preferences in Conjoint Experiments. Political Analysis, 1–15. https://doi.org/10.1017/pan.2019.30
León, S., & Orriols, L. (2019). Attributing responsibility in devolved contexts. Experimental evidence from the UK. Electoral Studies, 59, 39–48. https://doi.org/10.1016/j.electstud.2019.01.001
Levi, M., & Stoker, L. (2000). Political Trust and Trustworthiness. Annual Review of Political Science, 3(1), 475–507. https://doi.org/10.1146/annurev.polisci.3.1.475
Lewis‐Beck, M. S., & Stegmaier, M. (2007). Economic Models of Voting. In R. J. Dalton & H.-D. Klingemann (Eds.), The Oxford Handbook of Political Behavior. Oxford University Press. https://doi.org/10.1093/oxfordhb/9780199270125.003.0027
Lodge, M., & Taber, C. (2012). Three Steps toward a Theory of Motivated Political Reasoning. In Elements of Reason (pp. 183–213). Cambridge University Press. https://doi.org/10.1017/cbo9780511805813.009
Manin, B., Przeworski, A., & Stokes, S. C. (2012). Elections and Representation. In Democracy, Accountability, and Representation (pp. 29– 54). Cambridge University Press. https://doi.org/10.1017/cbo9781139175104.002
Matthew J Streb, B. B. (2008). Social desirability effects and support for a female American president. Public Opinion Quarterly, 72, 76–89.
Mauro, P. (1995). Corruption and Growth. The Quarterly Journal of Economics, 110(3), 681–712. https://doi.org/10.2307/2946696
McCann, J. A., & Domínguez, J. I. (1998). Mexicans React to Electoral Fraud and Political Corruption: an Assessment of Public Opinion and Voting Behavior. Electoral Studies, 17(4), 483–503. https://doi.org/10.1016/S0261-3794(98)00026-2
McDermott, R. (2002). Experimental Methodology in Political Science. Political Analysis, 10(4), 325–342. https://doi.org/10.1093/pan/10.4.325
Mestre, M. V., Samper, P., Frías, M. D., & Tur, A. M. (2009). Are women more empathetic than men? A longitudinal study in adolescence. Spanish Journal of Psychology, 12(1), 76–83.
192 Bibliography
https://doi.org/10.1017/S1138741600001499
Mishler, W., & Rose, R. (2001). Political Support for Incomplete Democracies: Realist vs. Idealist Theories and Measures. International Political Science Review, 22(4), 303–320.
Moore, A. B., Clark, B. A., & Kane, M. J. (2008). Who Shalt Not Kill? Individual Differences in Working Memory Capacity, Executive Control, and Moral Judgment. Psychological Science, 19(6), 549–557. https://doi.org/10.1111/j.1467-9280.2008.02122.x
Morton, R. B., & Williams, K. C. (2010). Experimental political science and the study of causality: From nature to the lab. In Experimental Political Science and the Study of Causality: From Nature to the Lab. Cambridge University Press. https://doi.org/10.1017/CBO9780511762888
Muñoz, J., Anduiza, E., & Gallego, A. (2016). Why do voters forgive corrupt mayors? Implicit exchange, credibility of information and clean alternatives. Local Government Studies, 42(4), 598–615. https://doi.org/10.1080/03003930.2016.1154847
Mutz, D. C. (2011). Population-based survey experiments. Princeton University Press.
Nye, J. S. (1967). Corruption and Political Development: A Cost-Benefit Analysis. American Political Science Review, 61(02), 417–427. https://doi.org/10.2307/1953254
Pancer, S. M., Brown, S. D., & Barr, C. W. (1999). Forming Impressions of Political Leaders: A Cross-National Comparison. Political Psychology, 20(2), 345–368.
Paparo, A., De Sio, L., & Brady, D. W. (2020). PTV gap: A new measure of party identification yielding monotonic partisan attitudes and supporting comparative analysis. Electoral Studies, 63, 102092. https://doi.org/10.1016/j.electstud.2019.102092
Pavão, N. (2015). The Failures of Electoral Accountability for Corruption: Brazil and Beyond. University of Notre Dame.
Pavão, N. (2018). Corruption as the Only Option: The Limits to Electoral Accountability. The Journal of Politics, 80(3), 996–1010. https://doi.org/10.1086/697954
193 Bibliography
Persson, A., Rothstein, B., & Teorell, J. (2013). Why anticorruption reforms fail-systemic corruption as a collective action problem. Governance, 26(3), 449–471. https://doi.org/10.1111/j.1468-0491.2012.01604.x
Peters, J. G., & Welch, S. (1978). Political Corruption in America: A Search for Definitions and a Theory, or If Political Corruption Is in the Mainstream of American Politics Why Is it Not in the Mainstream of American Politics Research? The American Political Science Review, 72(3), 974. https://doi.org/10.2307/1955115
Peters, J. G., & Welch, S. (1980). The Effects of Charges of Corruption on Voting Behavior in Congressional Elections. The American Political Science Review, 74(3), 697–708. https://doi.org/10.2307/1958151
Powell, G. B., & Whitten, G. D. (1993). A Cross-National Analysis of Economic Voting: Taking Account of the Political Context. American Journal of Political Science, 37(2), 391. https://doi.org/10.2307/2111378
Pring, C. (2017). People and Corruption: Citizens’ Voices from Around the World - Global Corruption Barometer.
Przeworski, A. (2018). Why bother with elections? Polity Press.
Reed, S. R. (1996). Political corruption in Japan. International Social Science Journal, 149, 395–405.
Riera, P., Barberá, P., Gómez, R., Mayoral, J. A., & Montero, J. R. (2013). The electoral consequences of corruption scandals in Spain. Crime, Law and Social Change, 60(5), 515–534. https://doi.org/10.1007/s10611-013- 9479-1
Rivero, G., & Fernández-Vázquez, P. (2011). Las consecuencias electorales de los escándalos de corrupción municipal 2003-2007. Estudios de Progreso, 59(201), 1.
Rosas, G., & Manzetti, L. (2015). Reassessing the trade-off hypothesis: How misery drives the corruption effect on presidential approval. Electoral Studies, 39, 26–38. https://doi.org/10.1016/j.electstud.2015.03.002
Rose-Ackerman, S., & Palifka, B. J. (2016). Corruption and government: Causes, consequences, and reform: Second edition. In Corruption and Government: Causes, Consequences, and Reform: Second Edition. Cambridge University Press. https://doi.org/10.1017/CBO9781139962933
194 Bibliography
Rothstein, B., & Varraich, A. (2017). Making Sense of Corruption. Cambridge University Press. https://doi.org/10.1017/9781316681596
Royzman, E. B., & Baron, J. (2002). The Preference for Indirect Harm. Social Justice Research, 15(2), 165–184. https://doi.org/10.1023/A:1019923923537
Rundquist, B. S., Strom, G. S., & Peters, J. G. (1977). Corrupt Politicians and Their Electoral Support: Some Experimental Observations. The American Political Science Review, 71(3), 954–963. https://doi.org/10.2307/1960100
Schleiter, P., & Voznaya, A. M. (2014). Party system competitiveness and corruption. Party Politics, 20(5), 675–686. https://doi.org/10.1177/1354068812448690
Schwindt-Bayer, L. A., & Tavits, M. (2016). Clarity of Responsibility, Accountability, and Corruption. Cambridge University Press.
Seidman, A. (2018, October 18). Corruption allegations? Voters don’t seem to care. The Philadelphia Inquirer.
Shleifer, A., & Vishny, R. W. (1993). Corruption. The Quarterly Journal of Economics, 108(3), 599–617. https://doi.org/10.2307/2118402
Singer, M. M. (2011). Who Says “It’s the Economy”? Cross-National and Cross-Individual Variation in the Salience of Economic Performance. Comparative Political Studies, 44(3), 284–312. https://doi.org/10.1177/0010414010384371
Sniderman, P. M. (2018). Some Advances in the Design of Survey Experiments. Annual Review of Political Science, 21(1), 259–275. https://doi.org/10.1146/annurev-polisci-042716-115726
Solaz, H., De Vries, C. E., & de Geus, R. A. (2018). In-Group Loyalty and the Punishment of Corruption. Comparative Political Studies, 52(6). https://doi.org/10.1177/0010414018797951
Stensöta, H., & Wängnerud, L. (2018). Why Expect a Link Between Gender and Corruption? In Gender and Corruption (pp. 3–20). Springer International Publishing. https://doi.org/10.1007/978-3-319-70929- 1_1
195 Bibliography
Sundström, A., & Stockemer, D. (2015). Regional variation in voter turnout in Europe: The impact of corruption perceptions. Electoral Studies, 40, 158–169.
Sundström, A., & Wängnerud, L. (2014). Corruption as an obstacle to women’s political representation: Evidence from local councils in 18 European countries. Party Politics, September(published online before print), 1–16. https://doi.org/10.1177/1354068814549339
Swamy, A., Knack, S., Lee, Y., & Azfar, O. (2001). Gender and corruption. 64, 25–55.
Tanzi, V. (1998). Corruption Around the World: Causes, Consequences, Scope, and Cures. International Monetary Fund, Staff Papers, 45(4).
Tavits, M. (2007). Clarity of Responsibility and Corruption. American Journal of Political Science, 51(1), 218–229. https://doi.org/10.1111/j.1540- 5907.2007.00246.x
Tay, L., Herian, M. N., & Diener, E. (2014). Detrimental Effects of Corruption and Subjective Well-Being. Social Psychological and Personality Science, 5(7), 751–759. https://doi.org/10.1177/1948550614528544
Teorell, J., Charron, N., Dahlberg, S., Holmberg, S., Rothstein, B., Sundin, P., & Svensson, R. (2013). the Qog Standard Dataset. University of Gothenburg, 504.
Tingley, Dustin; Yamamoto, Teppei; Hirose, Kentaro; Keele, Luke; Imai, K. (2014). mediation: R package for causal mediation analysis. In UCLA Statistics/American Statistical Association. Foundation for Open Access Statistics.
Torgler, B, & Valev, N. (2006). Corruption and age. Journal of Bioeconomics, 8, 133–145.
Torgler, Benno, & Valev, N. T. (2010). Gender and public attitudes toward corruption and tax evasion. Contemporary Economic Policy, 28(4), 554–568. https://doi.org/10.1111/j.1465-7287.2009.00188.x
Transparency International. (n.d.). Get involved - Report corruption. https://www.transparency.org/reportcorruption
Treisman, D. (2007). What Have We Learned About the Causes of
196 Bibliography
Corruption from Ten Years of Cross-National Empirical Research? Annual Review of Political Science, 10(1), 211–244. https://doi.org/10.1146/annurev.polisci.10.081205.095418
Uhlmann, E. L., Zhu, L. (Lei), & Tannenbaum, D. (2013). When it takes a bad person to do the right thing. Cognition, 126(2), 326–334. https://doi.org/10.1016/J.COGNITION.2012.10.005 van der Eijk, C., van der Brug, W., Kroh, M., & Franklin, M. (2006). Rethinking the dependent variable in voting behavior: On the measurement and analysis of electoral utilities. Electoral Studies, 25(3), 424–447. https://doi.org/10.1016/j.electstud.2005.06.012
Verba, S., Burns, N., & Schlozman, K. L. (1997). Knowing and Caring about Politics: Gender and Political Engagement. Journal of Politics, Vol 59(4), 1051–1072. https://doi.org/10.2307/2998592
Vidal, G. (2018). Challenging business as usual? The rise of new parties in Spain in times of crisis. West European Politics, 41(2), 261–286. https://doi.org/10.1080/01402382.2017.1376272
Villoria, M., Van Ryzin, G. G., & Lavena, C. F. (2013). Social and Political Consequences of Administrative Corruption: A Study of Public Perceptions in Spain. Public Administration Review, 73(1), 85–94. https://doi.org/10.1111/j.1540-6210.2012.02613.x
Villoria Mendieta, M., & Jiménez Sánchez, F. (2012). La corrupción en España (2004-2010): datos, percepción y efectos. Reis: Revista Española de Investigaciones Sociológicas, ISSN 0210-5233, No 138, 2012, Págs. 109-134, 138, 109–134.
Vivyan, N., Wagner, M., & Tarlov, J. (2012). Representative misconduct, voter perceptions and accountability: Evidence from the 2009 House of Commons expenses scandal. Electoral Studies, 31(4), 750–763. https://doi.org/10.1016/j.electstud.2012.06.010
Wagner, M., & Kritzinger, S. (2012). Ideological dimensions and vote choice: Age group differences in Austria. Electoral Studies, 31(2), 285–296. https://doi.org/10.1016/J.ELECTSTUD.2011.11.008
Wagner, M., Tarlov, J., & Vivyan, N. (2014). Partisan Bias in Opinion Formation on Episodes of Political Controversy: Evidence from Great Britain. Political Studies, 62(1), 136–158.
197 Bibliography
https://doi.org/10.1111/j.1467-9248.2012.01002.x
Waldmann, M. R., & Dieterich, J. H. (2007). Throwing a Bomb on a Person Versus Throwing a Person on a Bomb. Psychological Science, 18(3), 247– 253. https://doi.org/10.1111/j.1467-9280.2007.01884.x
Wei, S.-J. (2000). How Taxing is Corruption on International Investors? Review of Economics and Statistics, 82(1), 1–11. https://doi.org/10.1162/003465300558533
Weitz-Shapiro, R., & Winters, M. S. (2016). Can Citizens Discern? Information Credibility, Political Sophistication, and the Punishment of Corruption in Brazil. The Journal of Politics, 79(1), 60–74. https://doi.org/10.1086/687287
Winters, M. S., & Weitz-Shapiro, R. (2013). Lacking Information or Condoning Corruption: When Will Voters Support Corrupt Politicians? In Journal of Comparative Politics (SSRN Scholarly Paper ID 1641615; Vol. 45, Issue 4). Social Science Research Network. https://doi.org/10.5129/001041513X13815259182857
Xezonakis, G., Kosmidis, S., & Dahlberg, S. (2016). Can electors combat corruption? Institutional arrangements and citizen behaviour. European Journal of Political Research, 55(1), 160–176. https://doi.org/10.1111/1475-6765.12114
Zaller, J. R. (1992). The Nature and Origins of Mass Opinion. Cambridge University Press. https://doi.org/10.1017/cbo9780511818691
Zechmeister, E. J., & Zizumbo-Colunga, D. (2013). The Varying Political Toll of Concerns About Corruption in Good Versus Bad Economic Times. Comparative Political Studies, 46(10), 1190–1218. https://doi.org/10.1177/0010414012472468
198